Agisoft Metashape
Agisoft Metashape => General => Topic started by: Dmitry Semyonov on August 26, 2013, 04:14:32 PM
-
Pre-release version of Agisoft PhotoScan 1.0.0 is available for download.
This pre-release is considered as unstable and is not recommended for production use.
Please let us know if you have any problems using this version.
Please post problems and/or questions concerning other PhotoScan version in separate topics.
This topic will be removed when the final 1.0.0 version is released.
Note
Project files created with 1.0.0 version can't be opened with earlier versions.
EDIT:
PhotoScan 1.0.0 build 1795 was finally released. Thank you for useful feedback.
Please use the links on the main PhotoScan pages for download:
Agisoft PhotoScan Standard edition (http://www.agisoft.ru/products/photoscan/standard/download)
Agisoft PhotoScan Professional edition (http://www.agisoft.ru/products/photoscan/professional/download)
-
Thx Agi! ...very nice changes!
-
Thanks a lot ... Pscan rocks ! 8)
-
Where is now located the "Filter by Points" option that was located in the toolbar of the "Photos" pallete? This is a very useful function for quick editing rought point clouds. I can't also find the "Show depth maps funcion that was in the Photos pallete.
I was able to locate the "Filter by Markers" function using right mouse-click over the markers in the Workspace pallete.
These are very important functions in my opinion.
-
Hello RHenriques,
The Filter by Points command was moved to the context menu of the 3D view (simply right click anywhere in the main model view).
Depth maps are now discarded by default (they were replaced by a dense point cloud data). In case you really need to keep them, there is an option in the Preferences dialog, Advanced tab. Unfortunately there is no way to display depth map thumbnails in this version. Are they really so useful? The dense point cloud should provide a better visual representation of possible problems in the data set.
-
Sounds all extremely cool! Many thanks. Special thanks for the multitexturing.
-
Hi Dmitry,
very good and very useful new functions. I am currently playing around with it under Linux Mint 15, 64 bit, nVidia Driver Version 310.44.
Image set of 153 21 megapixel aerial images from Canon EOS 5d
Alignment, Scene reconstruction and dense point cloud in high mode work fine (240 Mio point).
But mesh generation fails, independent of settings, in the following way:
mesh generation dialog comes up after a short while, options can all be set as necessary.
When clicking ok, mesh generation starts, when progressbar reaches approx. 10-12%, progressbar dialog is closed and a mesh with 0 faces is generated. Thats it. Am I missing something?
Cheers
-
Thanks Dmitry. I'm now generating a dense point cloud from a previous project. I'll test things after. I'm curious about the orthophoto generation in this version since this is a very important function for me.
I think that it would be more logical to keep filter functions handy and not hide them only into context menus. Maybe you could keep a "filter by" button and add all the filter possibilities as a menu. This button should be in the left of the current "Reset filter" button. This way, in my opinion, is more logical than spread filter functions all over the interface (which is also good and you must keep them) without a centralized option to do so in the "Photos" palette. You could even use the "Reset filter" button without the red X over it. Options would be grayed on or off according to selection (For instance if a user selected points from the 3D view or a marker from the GCP view). Filter should include, at least, the options "By marker" and "By selected points".
Considering the depth maps thumbnails, they are not that useful indeed, considering the dense point cloud option. This is a great add to this version!
Keep on the great work!
Cheers
-
Hi,
Suggestions to make interfacing to other software more comfortable:
Import and export functions for dense point cloud / mesh to Point Cloud library PCD format and Cloud Compare Bin-Format, maintaining XYZRGBNXNYNZ attributes. Would save a lot of time that is otherwise used for intermediate conversion-steps.
Would be really great!
Cheers
-
Hi again Dmitry.
This version is very promising, at least for aerial photography processing. I've generated, from a previous project already optimized and with markers, a dense point cloud in medium quality. After, I've generated a mesh from the dense point cloud with the suggested high settings. The number of polygons calculated by PS (suggested value) was quite lower than the number calculated in the previous version mesh with polygon number defined as 100000000 to generate the maximum number possible of polygons. PS 1.0b gave me about 4 700 000 polygons and version 0.9.1 gave me 25 000 000 polygons. Even so, differences are almost unperceivable and, to be honest, it seems that the 4.7 million surface is more similar to the real world surface than the 25 million one. This is great if it is consistent for all projects. The good news here is that, with this new PS workflow, you can generate a great quality mesh in less time and much less RAM. I'm impressed with the results so far. DEM's also tend to be smaller for the same final terrain quality.
Considering the orthophoto generation, the color correction feature works great and the general quality of the orthophoto is better (sometimes, in some areas, much better) than the previous version. I've noted impressive improvements in houses rooftops (the dem also produces better results here) and in areas with moving vegetation. I've only noted that PS tends to ghost slightly some moving objects such as cars, but this is not a problem considering the overall improved quality of the orthophoto, which is, by far, more important.
Just one question. To generate the highest possible mesh quality we still can use custom mesh settings with a high polygon number, as in the previous version?
I'm still trying to figure out how the "Classify ground points" from the dense point cloud works. It seems a very nice and powerful feature, similar to LAS tools, that can give us control to filter a wide range of terrain features. Although you must provide some documentation to give us perfect control over it.
Cheers
-
Awesome work Agisoft!
A great software package that just keeps getting better!!! ;D
-
Hello Dmitry,
first of all, great improvements!!! Edge extraction of buildings in height field mode improved also a lot.
My questions:
What is the exact function of class assignment? I tryed to classify some objects as forest or building and started the mesh operation. The was no difference to the unclassified model. Is the behaviour of the mesh filtering different when working with classes? For me there was no difference. Or is it only to ignore classified objects during the mesh triangulation step???
As RHenriques mentioned, the maximal number of faces never reaches the max possible face count. It would be nice to have the possibility to set it to max in advanced mode. Mesh looks "denser" in the previous version (0.9.1). For example: from 8 000 000 points, i can create only 3 000 000 faces.
The recommended points per photo in version 0.9.1 was 40 000. Now the default setting is 5 000. So does it mean that the quality of the extracted features in alignment step is better (interest value is higher)?? This would speed up the algin step a lot :) :)
Edit: Seems like 5000 was my last setting in previous version, so it was taken for the new version too.
Cheers
Porly
-
Thank you for positive feedback.
photogrammetrix
Can you please provide a processing log (from the the Console pane) for the case with failed mesh generation?
RHenriques
Thank you for suggestions. We will consider them shortly.
photogrammetrix
While both PCD and Cloud Compare file formats are documented and can be implemented in PhotoScan, it is not clear what are the benefits of these compared to binary PLY file format.
It looks like both PCD and Cloud Compare support PLY format natively. Moreover, all these formats provide similar file sizes, and support point colors and normals equally well.
Have you tried PLY format with PCD or Cloud Compare? Are there any problems with it?
RHenriques
Thank you for feedback regarding ghosting in case of moving objects. We are going to investigate it in more detail.
The polygon count can be set to a custom value as before. Several presets with different polygon counts are provided mainly for convenience.
Documentation on the new functionality is still in the works. We will post the link here as soon as it is available.
Porly
Point classes are mainly used to exclude specific portions of the point cloud from triangulation step. This can be configured using Point classes option in the Build Mesh dialog (e.g. for DTM generation).
Besides that point class assignment has no effect on the generated surface.
The polygon count can be set to any desired value using Custom option.
The default points per photo value is still 40000. You may have changed it earlier, as PhotoScan doesn't reset settings to defaults after update.
-
Very cool! thank you Agisoft, will be testing.
From using the PS091_coordinates_to_bounding_box_rotate.py script which will correctly orient scenes to match 3rd party application orientations. I can see that predefined views are interpreted wrong.
As we would expect Z should represent front, back, Y should be top,bottom and x should be right,left
At the moment top, front, left views are wrong angles, and mapped incorrectly.
I'm also interested how we define model / scene pivot when using the new rotation tool?
-
Hello Infinite,
We have selected Z-up Y-front X-right convention for PhotoScan to be consistent with engineering data. Does it introduce significant problems in your case?
The pivot is currently not changed by a rotation tool, only orientation is affected.
-
Hello Infinite,
We have selected Z-up Y-front X-right convention for PhotoScan to be consistent with engineering data. Does it introduce significant problems in your case?
The pivot is currently not changed by a rotation tool, only orientation is affected.
Hi Dmitry,
Thanks for your reply.
That's pretty bizarre that in engineering terms Z-up Y-front X-right? generally in computer graphics, Z-Front (as in Z depth) X-right (as in horizontal) Y-up (as in vertical) xyz = rgb
It might be useful to have a tick box in preference to swap modes? It doesn't introduce problems it just appears incorrect.
The pivot is currently not changed by a rotation tool, only orientation is affected.
Doesn't the rotation tool rely on a pivot point to rotate around? if so what is the location of the pivot point in Agisoft?
- Lee
-
Hello Infinite,
We have selected Z-up Y-front X-right convention for PhotoScan to be consistent with engineering data. Does it introduce significant problems in your case?
The pivot is currently not changed by a rotation tool, only orientation is affected.
Hi Dmitry,
Thanks for your reply.
That's pretty bizarre that in engineering terms Z-up Y-front X-right? generally in computer graphics, Z-Front (as in Z depth) X-right (as in horizontal) Y-up (as in vertical) xyz = rgb
It might be useful to have a tick box in preference to swap modes? It doesn't introduce problems it just appears incorrect.
The pivot is currently not changed by a rotation tool, only orientation is affected.
Doesn't the rotation tool rely on a pivot point to rotate around? if so what is the location of the pivot point in Agisoft?
- Lee
No actually it's normal in engineering for Z to be UP, it's been like that for as long as I can remember.
"z-up was started with architecture in mind where x and y are the horizontal planes and thus z is the extrusion or height dimension. y-up was started with film in mind where x and y are the dimensions of the image plane and z is the depth or distance from camera."
There are other reasons as well... but this makes sense.
Cheers,
J
-
Hello Infinite,
We have selected Z-up Y-front X-right convention for PhotoScan to be consistent with engineering data. Does it introduce significant problems in your case?
The pivot is currently not changed by a rotation tool, only orientation is affected.
Hi Dmitry,
Thanks for your reply.
That's pretty bizarre that in engineering terms Z-up Y-front X-right? generally in computer graphics, Z-Front (as in Z depth) X-right (as in horizontal) Y-up (as in vertical) xyz = rgb
It might be useful to have a tick box in preference to swap modes? It doesn't introduce problems it just appears incorrect.
The pivot is currently not changed by a rotation tool, only orientation is affected.
Doesn't the rotation tool rely on a pivot point to rotate around? if so what is the location of the pivot point in Agisoft?
- Lee
No actually it's normal in engineering for Z to be UP, it's been like that for as long as I can remember.
"z-up was started with architecture in mind where x and y are the horizontal planes and thus z is the extrusion or height dimension. y-up was started with film in mind where x and y are the dimensions of the image plane and z is the depth or distance from camera."
There are other reasons as well... but this makes sense.
Cheers,
J
I really can't get my head around that x and y being the same plane. Very bizarre. It's like color, or colour. Left line, right lane :P or Unit measurements across different applications.
It would still be good to have this option. As you know Softimage z is depth, in Max y is depth. They always have to be different! so options for us confused and delirious users are always welcome!
-
I hear ya... that used to drive me bonkers as well (um...still does...LOL). But that's the reason Max uses Z up... it's a holdover from Autocad thinking.
Cheers,
J
-
I hear ya... that used to drive me bonkers as well (um...still does...LOL). But that's the reason Max uses Z up... it's a holdover from Autocad thinking.
Cheers,
J
Understood :D
-
Today I was able to generate a dense point cloud in a MacBook Pro, with only 8 GB of RAM, in "High" quality. This took a while but not as much as the mesh generation process, in medium quality settings, in the previous version. The posterior meshing process is impressively quick and the generated surface surpasses, by far, the best results I could achieve in the previous version. Much more detail and slightly higher accuracy. This move to allow the user to access and use the intermediate Dense Point cloud generation has proven to be one of the wisest decisions by Agisoft team toward quality. One of the best things for the user is the improved speed while tweaking the final model. With a modest configuration was almost impossible to edit the final mesh, mainly the dense ones. A simple selection of polygons could take several seconds. Making the same editing using the dense point cloud is a breeze. Users will like it a lot!
Cheers
PS: Considering the axis, z-up; y-front; x-right is the standard orientation in GIS systems. It's the standard for all the projected Earth coordinate systems. Allow them to change could be interesting for some external packages related with CAD applications but, for GIS users, PS's axis orientation is perfect as is.
-
I hear ya... that used to drive me bonkers as well (um...still does...LOL). But that's the reason Max uses Z up... it's a holdover from Autocad thinking.
I think you mean Y up is a holdover from 2D camera oriented thinking. :P
-
What is the meaning of "Image Quality"
How does it use?
Is it a feature which excludes the image of low quality (contribution) ?
-
PS: Considering the axis, z-up; y-front; x-right is the standard orientation in GIS systems. It's the standard for all the projected Earth coordinate systems. Allow them to change could be interesting for some external packages related with CAD applications but, for GIS users, PS's axis orientation is perfect as is.
I hear ya... that used to drive me bonkers as well (um...still does...LOL). But that's the reason Max uses Z up... it's a holdover from Autocad thinking.
I think you mean Y up is a holdover from 2D camera oriented thinking. :P
From now on and forever in computer graphics terms, I will always correct researchers and graphics artists when they use crazy terms like z-depth. "NO! NO! NO! I shall have you know it is called Y-depth! no X-depth!" Oh I give up :o
-
From now on and forever in computer graphics terms, I will always correct researchers and graphics artists when they use crazy terms like z-depth. "NO! NO! NO! I shall have you know it is called Y-depth! no X-depth!" Oh I give up :o
LOL!!! What a palaver ;)
Cheers,
J
-
I cannot get it to export an ortho as good as the last version. I have tried numerous settings and not had any luck.
Mainly in regards to vertical features, such as my vehicle. See attached samples...
-
Airmap3d, that still happens in my case but, mostly, only in areas where the number of projections is low (less than 3 photos). Things improved a lot considering geometric defined structures, such as homes.
From now on and forever in computer graphics terms, I will always correct researchers and graphics artists when they use crazy terms like z-depth. "NO! NO! NO! I shall have you know it is called Y-depth! no X-depth!" Oh I give up :o
From a GIS point of view, z is depth so z-depth is correct :)
To describe the surface os the earth, Y is used for Latitude, X for Longitude (geoidal surface of the earth after unfolding or projection) and Z for Altitude (or depth hehehehe), precisely as PS defines the Axis orientation when real world coordinates are used. In most 3D applications, such as Maya, Cinema 4D, you will get the same Axis orientation if you use the Front View. So, indeed, if you get a human figure orientated in front view standing up, X will be from hand to hand, Y from feet to head and Z will be the depth. The confusion is that, in Mapping and GIS applications, this is a top view and not a front view. It's indeed confusing and that's why it could be useful to allow axis shifting according to convenience.
This problem goes far beyond this because is one of the main issues when you try to import terrain models, generated from real world data, to most 3D applications. When we import a DEM, for instance, it's a mess. The terrain does not adjust to the planar default grid unless you do a X-90 degree shifting. Real world coordinates are all messed up. This is important if you need to use most dynamic and atmosphere effects from 3D apps. GIS and Mapping applications do not mix well with 3D modeling and animation applications unless you do some tweaks to your data, as I always do.
-
RHenriques, that particular portion of my ortho was projected in 15 images??? Both versions I ran with exactly the same data set and got significantly different results?
It's a mystery to me!
-
I really like the new improvements. Love the fact that now the software notifies you when trying to export and the photo folder is changed.
Also love the fact that we can see the project with dense point cloud because until now I could only look at the project with sparse point cloud or with mesh but that was moving so slow.
Did anybody noticed that now on the orthophoto we have ghost like objects when we have moving objects like cars? As a matter of fact RHenriques noticed that in the first place :)
On the leftt we have ortho using previous version and right using the new one. Notice that the color on the right one is much faded but I didn't use the color correction setting but the building geometry on the advertising panel is much better.
Also noticed the render info when exporting.
-
Very much enjoying this version, especially F2 to rename chunks, and the texture blending is astonishingly good!
Found that photoscan crashed on calculating colour correction if some photos weren't found and i chose to ignore the warning. (There were 3 missing out of 90 so i thought it would be ok, i know i need to be more organised!)
Is there somewhere I can run the photo location check without starting a workflow process, or other way to visually flag missing images, maybe in the photos pane?
-
OakCorp
Image quality can be estimated using "Estimate Image Quality" command from the Photos pane context menu. The quality values range from 0 (blurred images) to 1 (sharp images) and are displayed in the Photos pane in Detailed view.
PhotoScan doesn't use the quality estimates in any way, leaving the user to decide whether to disable blurred photos or not.
gheflorian
The ghosting effect in place of moving objects you noticed is a side effect of a completely new approach for orthophoto generation, which is capable of higher quality outputs than previously used one. We are working on possible solutions to minimize ghosting as much as possible.
James
Image check is currently performed only during processing steps. Usually the image paths become broken for all photos at once, so it is easy to check this by opening any photo. Of course this will not work if only few image paths are broken.
-
PhotoScan 1.0.0 distribution was updated to build 1736 today, which fixes major problems reported so far. If previous build didn't work properly for you - please update.
In case there are still any problems, please submit a crash report or post here. All reports are studied very carefully.
Thank you for constructive feedback and for trying latest PhotoScan version.
-
Here is what I have tested so far. It seems that indeed the buildings are constructed better and the colors are changed but the edges are not so straight as in sharp geometry anymore in some cases.
On the right is the example from the new version done with 1734 build.
-
here is another one
-
final one.
Sorry but file attachment sucks big time :)
-
Alexey, could be lifted attachment size ? to some "normal" sized images ? say to 500k ?? its pain to rescale images for the forum or uploading to some sharing sites.....
-
Alexey, could be lifted attachment size ? to some "normal" sized images ? say to 500k ?? its pain to rescale images for the forum or uploading to some sharing sites.....
+1 to this! ;)
-
Max attachment size was changed to 500k.
-
Max attachment size was changed to 500k.
thank you!
here are some other comparison photos!
-
as you can see model reconstruction is much better in almost every situation.
-
Hello gheflorian,
very useful for comparison.
Left is alwyas 1.0 and right 0.9.1???
-
Hello gheflorian,
very useful for comparison.
Left is alwyas 1.0 and right 0.9.1???
Hello Porly,
Sorry left is for 0.9.1 and right is 1.0 :)
The posts from today are from the latest build 1736.
-
as you can see model reconstruction is much better in almost every situation.
That might be so for aerial photography and reconstruction but there is a drop in quality with multi-camera body scanning, due to dense point cloud integration. Which I have flagged with Agisoft team.
The new mode is great for editing point clouds but the results are softer and we loose details.
Example here:
http://www.ir-ltd.net/uploads/agi-nodense01.jpg
http://www.ir-ltd.net/uploads/agi-dense01.jpg
http://www.ir-ltd.net/uploads/agi-nodense02.jpg
http://www.ir-ltd.net/uploads/agi-dense02.jpg
This is not ideal as we want to be improving detail, not loosing it.
-
And i?m was thinking seeing bad :-)) thanx Lee.....
-
Hi Infinite. Does the softness really has something to do with the dense point cloud integration? As a matter of fact, from what I've learned, the dense point cloud generation has always been an intermediate step in mesh generation, even in earlier PS versions. The problem is that we weren't able to access to it. Did you check the number of polygons or any other detail that might be responsible for the difference in models? This new workflow is great because we can get much better results with more modest configurations. In my portable, for instance, I was not able to pass "medium" quality meshes and it took ages to render. Now I can get, at least High quality point clouds and the results are a lot, lot better than earlier versions and a lot faster to obtain. It's precisely the opposite of your example - meshes have a lot more detail, sometimes breathtaking, and prominent geometric objects, such as houses and similar objects, are much more similar to the real world. But, as you said, I'm using Heightfield and working with aerial photography. Maybe there are some differences in the Arbitrary option that you are certainly using as preferred option.
Cheers
-
Hi Infinite. Does the softness really has something to do with the dense point cloud integration? As a matter of fact, from what I've learned, the dense point cloud generation has always been an intermediate step in mesh generation, even in earlier PS versions. The problem is that we weren't able to access to it. Did you check the number of polygons or any other detail that might be responsible for the difference in models? This new workflow is great because we can get much better results with more modest configurations. In my portable, for instance, I was not able to pass "medium" quality meshes and it took ages to render. Now I can get, at least High quality point clouds and the results are a lot, lot better than earlier versions and a lot faster to obtain. It's precisely the opposite of your example - meshes have a lot more detail, sometimes breathtaking, and prominent geometric objects, such as houses and similar objects, are much more similar to the real world. But, as you said, I'm using Heightfield and working with aerial photography. Maybe there are some differences in the Arbitrary option that you are certainly using as preferred option.
Cheers
This was a direct comparison. No difference other than older Agisoft build and latest Agisoft build. I've often seen changes in quality during each release, some good some bad. This one the results are softer.
There are however some interesting speed boosts. Which is a good thing! don't get me wrong the results are still fantastic but we should always be pushing for a raise in quality as much as possible.
-
I really liked the 'filer by markers' button that was on the photo pane. Any chance of putting that back on there?
-
Hello, i am getting this errors in 1.0 (last build) which i dont have in 0.91 build 1703 (both 64bit).
clGetDeviceIDs failed: CL_DEVICE_NOT_FOUND
Intel(R) OpenCL: can't enumerate devices
My specs:
OpenGL Vendor: NVIDIA Corporation
OpenGL Renderer: GeForce GTX 470/PCIe/SSE2
OpenGL Version: 4.3.0
Maximum Texture Size: 16384
Quad Buffered Stereo: not enabled
ARB_vertex_buffer_object: supported
ARB_texture_non_power_of_two: supported
However, everything works well so far.
Admir
-
After some few tests with aerial photography, 3D models are consistently better than in earlier versions. Although, considering the orthophoto generation, results are mixed. Color correction works great and it's an added value for quality but, as with airmap3d, I've also got some warped features that were better in the previous version. The problem in that I've also got some features that were warped in the previous version and are now great. The final result is better in this version but a better effort must be made to avoid such warped features. There might be some stochastic or random approach in orthophoto generation because I'm obtaining different results from the same project with slightly tweaked parameters in the dense point cloud (mainly filtering). Even if I keep everything at the same state, othophotos get slightly different in two exportation tries. Even so, considering all the pros and cons, the final results are better than in the previous version.
Airmap3D, you can now click over a marker from the GCP pallete to filter the photos. You can also filter by markers by right-click in the model's window. Even so, as you said, I also like more to have that filtering in the photos pallete too, as in the previous version. If you check my first post in this thread, you will see that your request was also my request when I first tried this version ;-)
Cheers
-
Is there some explanation about the new Build Texture parameters? What means texture size / count? I've check the guide and the changelog, but I can't find nothing about it...
-
Hello, i am getting this errors in 1.0 (last build) which i dont have in 0.91 build 1703 (both 64bit).
clGetDeviceIDs failed: CL_DEVICE_NOT_FOUND
Intel(R) OpenCL: can't enumerate devices
My specs:
OpenGL Vendor: NVIDIA Corporation
OpenGL Renderer: GeForce GTX 470/PCIe/SSE2
OpenGL Version: 4.3.0
Maximum Texture Size: 16384
Quad Buffered Stereo: not enabled
ARB_vertex_buffer_object: supported
ARB_texture_non_power_of_two: supported
However, everything works well so far.
Admir
I've the same card and it work perfectly with 1.0.0 1736. Try with the latest release of GeForce Driver (326.80).
-
DJBill, it create more than one UV map-textures, that mean on bigger project you must not to use 16k+ textures, that often problem to properly view or process.... in short, GREAT addition to texture quality mostly on BIG datasets....
-
Thank for the reply... I'm currently test this new feature with 1240 cameras (inside part of an old tower) with this settings: 4096 x 16. I hope it works ;)
-
The new Build Texture works very good, but the bundle adjust pass takes so much time... :S
-
Hi,
In the "chunk merge" part, in version 0.9.1 the mesh is merged to one closed surface, but in this new version, the mesh is combined to several layer, the results are totally different. And I really want to keep the old merge, or give two choice here: merge mesh or combine mesh.
-
Hi,
it's a really good think to make the build dense cloud app, it can be very usefull BUT,
i'm disappointed there is not a red/green mask for the picture to texture. It can be very usefull to me to draw a "green zone" and force photoscan to use this mask as a texture.
Maybe for the next step, i hope
-
Hello Albert,
In the previous version geometry merging procedure produced low quality geometry irrespectively to the quality used in original chunks. In the version 1.0.0 you can merge chunks without geometry and build mesh in the merged chunk already.
-
Hello Albert,
In the previous version geometry merging procedure produced low quality geometry irrespectively to the quality used in original chunks. In the version 1.0.0 you can merge chunks without geometry and build mesh in the merged chunk already.
absolutely awesome ... 8)
-
seems like "Assign class" has not been activated yet ...
-
Hello George,
Assign Class feature is active in case you have selected at least one point of Dense Point cloud (and surely it will not work for Sparse Point cloud selection).
-
aha! got it!
but how can I select? With the common tool kit in the upper ribbon, right?
-
Hello George,
Yes, the same selection tools are valid for dense cloud. Please also note that there are two view modes for dense cloud: colored by object colors (similar to sparse point cloud view and shaded mesh mode) and colored by assigned classes.
-
finally made a brief test
seems like classified Dense point cloud and corresponding options for mesh generation etc. are fantastic features.
I've tested PS 1.0.0 at my latest dataset covering 450 ha of a forested area (2500 images) and generated DTM (quasi one) from DPC; and I compared the result with official DTM (produced based on LIDAR of just 15 cm accuracy). What I have got is very "little" relative difference. Not exceeding 1,5 meters (an average is around 1 meter) at more or less sparse forest and less than 0,2-0,3 meters at open areas. When it comes to a dense forest there are still issues but looks promissing. Dense forest is actually never easy surface. But guys this is 450 ha or 4,5 sq.km - quite a large surface tho.
-
The user reports regarding 1.0.0. seems very promising with regards to functionality and performance.
Does this new version somehow impact the recommended / required hardware profile?
(I know there is wiki info on hardware matters but that info is based on 0.9.1 and I figured the new version seems to perform differently?)
I will shortly invest in a workstation and I am looking for the sweet spot with regards to performance & capacity / price.
-
Hello mr map,
Actually, hardware recommendations are the same for version 1.0.0.
-
Hi,
anyone could tell me because the bundle_adjust pass takes a couple of hours with an AMD Phenom X6 1055T and 16GB of RAM with a set of 1240 photos? I don't remember this pass during texture creation in 0.9.1...
Thanks in advance
-
Hello djbill,
It's due to Color correction feature. If you switch it off in the Build Texture dialog the processing time will be reduced significantly.
-
OK, thank you so much for the reply. Color correction feature works very well, it's a pity doesn't use it ;) I want just understand why it take long time.
-
Question for Alexey...
Will there be a new user manual / tutorials that accompany the new version once it is officially released?
Cheers,
Airmap3D
-
Hello Airmap3D,
Of course, there will be new manuals and tutorials for the version 1.0.0. Currently we can't publish finished version, since there are frequent changes in GUI and functionality and in this case such materials could be outdated in a few weeks.
-
Hello again. I have been testing 1.0 build 1736 (x64) and getting much worse align than in 0.91 buld 1703 (x64). Both alignment processes have same settings: High, no image pair preselect, max points is 80000 and mask constrain is ON. Attached is shot from 1.0 and 0.91.
-
Hello admir,
Could you please provide the dataset where we can reproduce alignment problems in 1.0.0 and good results with 0.9.1?
-
Sure, you need photos, masks and projects ?
Hello admir,
Could you please provide the dataset where we can reproduce alignment problems in 1.0.0 and good results with 0.9.1?
-
Hello admir,
At least we need photos and masks, but project files (just with the result of alignment) will be also useful for comparison.
-
helluuu
Unless I am blind.. I can't find View UV's anymore..... 8) this might come in handy sometimes
-
Hello FoodMan,
Tools -> Mesh -> View Mesh UVs
-
ahah thanks Alexey.. indeed I am blind ;D
f/
-
Hello Albert,
In the previous version geometry merging procedure produced low quality geometry irrespectively to the quality used in original chunks. In the version 1.0.0 you can merge chunks without geometry and build mesh in the merged chunk already.
1. In 1.0.0, I can merge chunks without geometry, but lost the dense point cloud, which is calculated in each chunk. I want to do like this way: in each chunk, calculate camera pos, sparse point, and dense points, then edit the dense point. After that, align and merge the dense point to a new chunk, finally get mesh based this dense point. Now I just can do this: calculate the cam pos and sparse point, then merge them, next calculate dense point, and mesh. This is bad for me, because in my app, some photos are global zoom out, (whole body) some are detailed zoom in (face). The zoom out photo will destroy the detailed photo.
If we merge two chunks (with dense pont, without mesh), that is not good to lost the dense point.
2. I see, in last version, if merge the mesh, will lost some quality, but can get a whole closed model, this is very important I think. In the new combined version, as I said before, if can combine the dense point, that is also good, but now only can combine the sparse point or mesh, for the mesh, it is not closed model, with several layer, this is really bad, I need to do a lot work to edit it. Why not give two possibility here, for some users, like me, we still need. :)
And thanks for your feedback.
-
Nice work AGISOFT Team on BETA-version 1.0.
Will there be a funtion in 1.0 to import a Dense Cloud in .las format ?
(which has been classified in an external Software)
Or can you give me a little hint , for a work around?
It would be nice to classify the Dense Point Cloud in an external Software aswell.
Or even if i erase some Dense Cloud Point unintentionally, i can restore them from an backup file (inclusive the classification), without creating the Dense point cloud again, and the classification.
Saving and loading of an whole big Project in PS takes too much time.
Maybe a "save-classification-mask" of Dense Point Cloud could be of a little help.
thx a lot.
-
wonderful program has become even better.
Great job developers.
-
delighted by this version
see attached:
all surfaces
ground
Almost like a LIDAR (not exactly but still very good move towards)
Please-please guys - resolve the ghost effect
Cheers
-
...
-
Hello Albert,
After merging two chunks you can re-build dense point cloud in the Merged Chunk and generate mesh based on this data. Also it's possible to create mesh based on sparse point cloud.
Hello 3DFranz,
We are planning import dense point cloud feature.
-
Hello Albert,
After merging two chunks you can re-build dense point cloud in the Merged Chunk and generate mesh based on this data. Also it's possible to create mesh based on sparse point cloud.
Hi,
If I re-build the dense point cloudin the merged chunk, will lost a lot of good detail information, and this detail surface can be got in each chunk separated. If the software can combine the dense point cloud, not simple sparse point cloud, that's perfect for me.
Or keep the merged tool in the last version. Thanks.
-
Hello Albert,
We'll try to implement dense point cloud merging soon.
-
Hi,
I've started processing some projects that were originally aligned and mesh built in 0.9.1. I achieved a mesh of 100,000,000 triangles from over 1300 photos, but there was a tiny error in a small portion of the mesh. Because 1.0.0 was released around then, I decided to reprocess using the dense point cloud and then mesh build with the same triangle count. I used the same computer (and tried it on two others as well) and found that the 1.0.0 pro version ran out of memory for every build I attempted. Even when I reduced mesh resolution count to 2/3 the original model. Any chance you could tell me if I'm doing something wrong? I really like the other features of the pre-release and would like to keep using it if I can get it to build to the same resolution I'm accustom to from the previous version.
Thank you!
-
Hello ksimon,
What quality for depth maps was used in 0.9.1 and in 1.0.0? Are you sure you've used the same parameter in both cases?
Probably you can also provide the cpoy of log file for mesh generation step.
-
Hey guys,
I'm evaluating PS Pro for orthophoto generation of forests and fields of crop-like vegetation. I've successfully rendered 3 orthophotos from 700-2700 15Mpix images (50-100ha), each in under 2-3 non-stop PC work-days. I started a fourth project from 2600 photos with this new version of PS. "Align photos" finished within a reasonable amount of time, then I chose a new step "Build dense point cloud". First phase of this step took about ~30h, and now "Generating dense point cloud..." estimates ~310h left. Since I'm quite a noob in this area, I dare to ask: can I skip this step, and move right to "Build mesh" as I'm used to? If I do so, what do I loose, and what do I gain, in terms of processing time and quality? I know I can just try to, but I'll loose those 30h of previous rendering and I have no other machine to try it on in parallel.
Thanks in advance for your input.
P.S. I don't do manual modification of intermediate results, nor I intend to.
P.S.S. My machine is i7@3.8GHz, 16MB RAM, GeForce GTX 660.
-
Unfortunately, this version is not stable.
In the final seconds of mesh generation. (Dense cloud)
The demand for memory.
Follows the jump by 200% and
And the work goes to waste.
A few projects I crashed in the final seconds.
-
small project 240 photos. Saving the project takes 14 minutes.
Is that a joke?
Does anyone sees any advantages of this version?
Dense cloud, less stable.
The inferior quality, fade detail, lack of details in the model.
The only progress is a multi textures.
ps.
Where can I download older versions?
-
Kris, go in TOOLS-PREFERENCES-ADVANDEC- and set compression level to lower number - for speedier save. it depends on how much data is in POINTCLOUD because its biged as with DEPHT maps. Stability is OK, had no crash with it, and im processing mesh up to 50mil trias
Yes as Lee pointed out, we loose a bit of details, but it still just BETA and Alexey and his team has a good reason why it goig to use PClouds, you will see it later when is all implemented....
The main application target for PSCAN is a GIS stuff, therefore its primary target group of users.... and its about deepr knowlege how to get better details out sometimes.......
-
Hello kris3d,
Project save could be long in several cases:
- saving via network,
- saving depth maps in ultra quality (in 1.0.0 you can disable saving depth maps in the project file using the corresponding option in the Advanced tab of PhotoScan preferences),
- saving dense point cloud with hundreds of millions points,
- saving undecimated mesh with hundreds of millions faces.
As it was already mentioned, you can turn off project compression in the Preferences window, thus minimizing save time, however, the project file will be huge.
Please provide additional information about the problems with the pre-release so we could check them.
The actual release (0.9.1) can be downloaded from the products' page: http://www.agisoft.ru/products/photoscan/professional/download/
-
Hello Rokas,
Please specify the reconstruction settings used for dense point cloud generation.
Please also check the current memory consumption in Task Manager.
-
before recording project lasted 30 seconds
-
Kris, if you happy with the MESH reconstruction, save it under clear naming convention so its clear it contain DEPHT and PCloud. Then delete DEPHT and PC and use just MESH. and as im see you have COPYs of same CHUNK so the size going up dramatically. try save EACH chunk separately and work this way.....
-
Tree definition noticeably improved with new point cloud system.
-
Using Build 1742 on a Linux Station 64 bits, I can't produce a mesh using the dense cloud as input data. I keep getting an error message "Can't build model".
This happens whatever project is used.
the procedure I use is the following :
1- align photos
2- build dense cloud
3- try to mesh dense cloud.
-
Hello Jules,
Could you please provide a copy of log file information related to this process? Please also check that Bounding Box is specified correctly.
-
here is the log file, the bounding box looks OK.
-
Hello Jules,
Could you please send us the project file?
-
as the problem is generic, I can send you any project.
link to the sample monument project :
https://upload.osupytheas.fr/qpv5fo41
-
Hello Jules,
And what are the settings used in the Build Mesh dialog?
We've tried this project file and do not have any problems neither on Wiindows nor on Linux.
-
Documentation on the new functionality is still in the works. We will post the link here as soon as it is available.
Looking forward to this!
-
Deleting part of the dense cloud does not make a change in the total point cloud. Does it mean that the full dense cloud is still used for generating the mesh or is it make use of the selection ? I'm asking that because when I try to generate a mesh it always run of memory, even when I set it up for only 5.000 faces (8go available).
-
Hello JRM,
Mesh is generated based on dense point cloud, if you delete some points from cloud - they will be no longer used for mesh generation (please note, that point removal from sparse cloud will not effect on mesh reconstruction if dense cloud has been already reconstructed).
Could you please provide a copy of log file related to mesh generation where memory problems appear?
-
Thanks for the answer. The log file is the output of the console tab or is it another thing ?
Is there a way to export the dense cloud instead of the sparse one ?
If you a .ts file it'll be a pleasure to translate the app in french.
-
Hello JRM,
Yes, you can just copy information from the Console pane.
To export dense point cloud you need to select Dense type in the export point dialog, providing that dense cloud has been already reconstructed for the chunk.
Please contact info@agisoft.ru if you are interested in translating PhotoScan GUI.
-
Here is the log (test done with Fedora 19)
-
Hello JRM,
Thank you for providing the log information. What is the amount of RAM and swap available on your computer?
-
4.2Gio of swap and 8Gio of RAM
-
Hello Alexey,
I have tried any combination of parameters. but here is a snapshot of the standard parameters and the error message.
to complete my post, if I choose Height Field instead of Arbitrary then I have no error message but I get an empty model.
-
Hello Jules,
I've managed to reproduce such behavior for Arbitrary and Height Field, but I've just moved Bounding Box away from dense cloud. Could you please check one again, that working volume is defined correctly and there is at least part of dense cloud inside the box.
-
Hello,
I had generated Medium quality depth maps in two chunks in 0.9.1. I can't remember if I merged the two chunks in 0.9.1 or 1.0.0, but there is no depth map included in the my merged chunk that resulted in very low resolution even though my dense point cloud was 42 000 000+ and I specified an 80 000 000+ face count in the geometry build. Please find the log file attached.
Similarly, my colleagues have found a variety of issues memory limits and low res results in geometry builds that are executed in 1.0.0 on projects that were aligned, etc in 0.9.1.
Thank you!
Katie
Hello ksimon,
What quality for depth maps was used in 0.9.1 and in 1.0.0? Are you sure you've used the same parameter in both cases?
Probably you can also provide the cpoy of log file for mesh generation step.
-
Hello Katie,
I meant, if you could attach the log file that corresponds to the insufficient memory problem.
I had generated Medium quality depth maps in two chunks in 0.9.1. I can't remember if I merged the two chunks in 0.9.1 or 1.0.0, but there is no depth map included in the my merged chunk that resulted in very low resolution even though my dense point cloud was 42 000 000+ and I specified an 80 000 000+ face count in the geometry build.
Without depth maps and dense point cloud calculated for the merged chunk you can't build high quality model, only sparse point cloud based geometry option is possible.
-
I'm sorry, I'm confusing two approaches I took to generate a new mesh in 1.0.0
In the first approach I ran out of memory in the mesh build and do not have the log file, but will replicate the scenario and send it. In the second approach I went back a step and tried to re-merge two chunks which were created in 0.9.1 in 1.0.0. This newly generated mesh does not preserve the depth maps from the initial two chunks and the mesh build was very low resolution. Do you know what I can do to preserve the depth maps in the chunk merge if this is limiting my mesh build resolution? The chunks say medium depth map and the properties show an aggressive depth filtering which was the setting in all process I've done.
Thank you!
Hello Katie,
I meant, if you could attach the log file that corresponds to the insufficient memory problem.
I had generated Medium quality depth maps in two chunks in 0.9.1. I can't remember if I merged the two chunks in 0.9.1 or 1.0.0, but there is no depth map included in the my merged chunk that resulted in very low resolution even though my dense point cloud was 42 000 000+ and I specified an 80 000 000+ face count in the geometry build.
Without depth maps and dense point cloud calculated for the merged chunk you can't build high quality model, only sparse point cloud based geometry option is possible.
-
Is it possible to install the beta separately and keep the 0.9.1 installation for comparison purposes?
-
Hello Katie,
Merge mesh option in the version 0.9.1 resulted in low quality mesh irrespectively to the geometry quality in the original chunks due to the methods used. In the version 1.0.0 Merge Mesh option was changed to Combine Mesh that produce single geometry from all the chunks being merged without any changes.
In both versions it is possible to merge chunks without any geometry reconstructed and build geometry in merged chunk. We are planning to implement dense cloud merging to the merge chunks feature.
-
Alexey,
yes I'm sure of the bounding box, sorry about it.
Hello Jules,
I've managed to reproduce such behavior for Arbitrary and Height Field, but I've just moved Bounding Box away from dense cloud. Could you please check one again, that working volume is defined correctly and there is at least part of dense cloud inside the box.
-
After a short glance and two tries I am totally enthused! Very well done! :)
-
The new dense point cloud functionality is great. Especially being able to edit it.
One request - I find myself switching back and forth between editing (usually free-form polygon) and navigating around the model with the arrow/navigate button. Would it be possible to add single-key shortcuts (awesome if they were customizable) so we could toggle between navigation and edit modes without point-clicking? This would save me a lot of time in point cloud editing mode.
-
Hi Alexey,
Can you elaborate/describe on the settings/colours for the dense cloud classifications. I cant find any mention of it in the forum/instructions.
Many Thanks
-
Hello andyroo,
To switch from Navigation mode to the last used selection tool in the Model view just press Space on keyboard.
-
Hello Matt,
Automatic dense point cloud classification finds points from the two classes: Ground and Low Points (that are under the estimated ground level). Manual classification allows to assign classes to the selected points.
It's recommended to use default values in Classify Ground Points dialog, but they can be slightly varied depending on the project.
Generally the method is the following: at first in every cell the lowest point is selected (Cell Size should be adjusted so that there are guarantied ground points in each cell). Then these points for each cell are triangulated in grid that should not be laying above the ground. Then new points are added to this grid - each of the points being added should match the following criteria:
the distance between the point and the "ground surface" should not be bigger than Max.Distance parameter; the angles between the "ground" and the vectors connecting the point and vertices of the corresponding cells are not greater then Max.Angle parameter.
Later in the Build Mesh dialog you can choose classes that should be used for mesh generation.
-
Is it possible to install the beta separately and keep the 0.9.1 installation for comparison purposes?
Good question??? I would like to have both on my computer so that I can continue to practice with V1.0 while still processing commercial jobs with V0.9, rather than having to un-install/ re-install every time.
-
Is it possible to install the beta separately and keep the 0.9.1 installation for comparison purposes?
Good question??? I would like to have both on my computer so that I can continue to practice with V1.0 while still processing commercial jobs with V0.9, rather than having to un-install/ re-install every time.
You can install 0.91, copy it to folder like PhotoScan_0.91, uninstall (it will remove it from Photoscan folder), install 1.0 (in Photoscan default folder). You end with 2 versions in 2 folders.
Admir
-
Just a few quick things:
I would just like to say 'thank you' for the new hotkey to switch between perspective and orthographic views. It would be nice to have others to switch between some of the tools (rotate and scale region are two I can think of)
Oh, and the editable dense cloud before building geometry is awesome...it will be a big time saver.
Cheers, -C
-
Am playing with the Classify Ground points tool (which is fantastic addition as have been tearing my hear out trying to get LAStools to work). Anyway I ran with defaults and worked pretty well. Then I played with settings and buildings were then classed as brown colour (same as the ground - ie the whole point cloud was brown). Then when I changed the settings back to default and reran nothing changed (everything remained brown). Didn't seem to be able to undo the previous classification. Am I missing something? Thanks
-
Hi does someone have samples of after classifying canopy that can be shown here in a jpg form?
-
Hello choppedxs,
In the Classify Ground Points dialog you can select if classification is applied to any class or to non-classified point only (please check that you are using build 1742).
-
Hello friends,
is there a way to skip the dens cloud generation step? It would be interesting for someone who dont need to edit anything in the dense point cloud. For aerial projectas the dense pointcloud can consist of one billion points, so editing wouldnt be possibile any more due the amount of data. Generating dense point cloud is a additional step in version 1.0. that can take several hours. It is an amazing option/improvement and real useful for smaller projects, but for most of my projects I would like to skip this procedure.
I also tried to generate depth maps seperately for all photos and go directly to generating mesh step (as in version 0.9.1), but I couldnt find the "build depth" option for each image in "Photo pane".
Maybe I missed something??
Thanks a lot!
Best regards
Porly
-
Hello Porly,
Actually, in previous versions dense point cloud was also generated from the depth maps, but this process was a sub-step of Build Geometry stage. So in 1.0.0 Build Dense cloud and Build Mesh will produce the result of Build Geometry stage in 0.9.1.
-
all (as far i know) photogrammetric solution work with pointclouds and then the PC is converted to triangles-mesh. Then depend on how is the meshing algorithm implemented AND optimized for various tasks. For now is the POISSON mesh reconstruction used as one of the best pros/cons from various techniques.
-
Sort of a feature request, but since it's related to the dense point cloud editing I thought I'd post it here. Feel free to move this post if needed.
Would it be possible to enable the Invert, Grow and Shrink Selection functions to Dense Point Cloud editing? It would be very helpful when trying to clean up the dense point cloud before building the mesh. I know it's not really the same function as with connected pieces of built geometry, but I thought I might as well ask.
Thanks! -Chad
-
Haven't had much opportunity to play with the beta, but I see the view presets made it in. :)
Really looking forward to some documentation on the new features!
-
Sort of a feature request, but since it's related to the dense point cloud editing I thought I'd post it here. Feel free to move this post if needed.
Would it be possible to enable the Invert, Grow and Shrink Selection functions to Dense Point Cloud editing? It would be very helpful when trying to clean up the dense point cloud before building the mesh. I know it's not really the same function as with connected pieces of built geometry, but I thought I might as well ask.
Thanks! -Chad
I second that. Or even just a 'decimate dense point cloud' function of sorts?
-
Hello Alexey,
just to inform you that my problem in building the mesh was due to locale settings. I had french locale numeric wich implied a mess in the coordinates and then in the BBox.
Thanks for you support,
regards
Hello Jules,
I've managed to reproduce such behavior for Arbitrary and Height Field, but I've just moved Bounding Box away from dense cloud. Could you please check one again, that working volume is defined correctly and there is at least part of dense cloud inside the box.
-
Hello Jules,
Thank you for informing.
-
Hi, I have a question.
In which point of process is there Classify Ground Points dialog? Is it in Build Dense Cloud or in Build Mesh? Because after Build Dense cloud I can see only white points without classification. Do I have to do it manually in some special step (Manual classification)?
Thanks for response.
-
Hello onder,
It's possible to classify ground points automatically only for georeferenced chunks.
-
Thank you for quick response.
-
Photos don't filter with selected marker.
Guided marker placement not very accurate requiring adjustment of nearly all markers. Manual method placing markers on two photos accuracy as expected (judging by Error (pix)).
-
Hello bisenberger,
To filter photos by markers now you need to select marker (or markers), right-click on it in the Workspce pane or in the Model view and choose the corresponding option from the context menu.
Guided approach may be inaccurate due to reconstructed geometry.
-
How long should "Colour Correction" take when exporting an ortho?
I have 1103 images (UAV)
Exporting a jpg at 10k max pixel dimension
Without colour correction enabled it exported in about a minute or two, with colour correction enabled it took about 5 minutes for the 'calculating colour correction' status to reach 100% and now nothing has happened for over 40 mins. Overall progress is stuck at about 25%.
Is this right?? Or has something gone wrong / Have I done something wrong?
-
Update...
At about 43 minutes it started the next step (rendering mosaic).
Question still stands though... Does it normally take that long for the colour correction?
-
Hello airmap3d,
Unfortunately, color correction really takes a long time for big datasets.
-
Hi Alexey,
That is fine. I just wanted to make sure that I hadn't got something wrong. I was just a little confused as the progress bar sat at 100% for about 30 minutes.
Thanks.
-
Along the same lines - time to complete a task - I have a similar question. I am looking at a dense point cloud reconstruction (1736 build) that I selected at medium resolution (1447 12 MP images). Right now PS says 27 hours elapsed (about right) and 91 hours remaining. I was surprised at how long it looks like this will take, since I picked medium resolution for both the alignment and the dense pointcloud. I am guessing I will end up with a huge number of points...
The images have a great deal of overlap - the point cloud was derived by attaching a camera with intervalometer to crane and lowering it into a canyon below a dam. Camera is looking downward and slightly obliquely. The crane moves slowly relative to the image frequency (1 image/3 seconds).
System is a dual CPU workstation with two quad core Xeon X5647 (2.93 GHz) CPUs, 192 GB 1333 MHz RAM and a single ATI HD7970 GPU.
-
Hello bisenberger,
To filter photos by markers now you need to select marker (or markers), right-click on it in the Workspce pane or in the Model view and choose the corresponding option from the context menu.
Guided approach may be inaccurate due to reconstructed geometry.
Thanks Alexey,
Applying the filter works like a charm... now how do I remove the filter?
Yes, the reconstructed geometry was based on the sparse point cloud with either the medium or low polygon count.
Is there a way to change the coverage area units from sq km to sq m when generating a report?
-
Hello bisenberger,
There's "Reset filter" button on the Photos pane.
Report files are not currently customizable.
-
Hi again,
Could you please explain how the different Build Mesh decimation levels are determined? I see that there are the Low, Medium and High options with their resulting poly amounts listed (along with the old school 'Custom' option); but how are those various levels calculated from the Dense Cloud stage?
thanks! -Chad
-
Another followup on my dense point cloud processing time observation above.
I decided to process a similar sized dataset (but from an aircraft covering 30 sq km instead of a crane covering ~10,000 sq m) and the dense point cloud processing finished in <6 hours. I am guessing that the increased overlap on the project I mentioned earlier is the main cause for the increase in processing time, as opposed to the total number of images.
(I also updated to build 1742)
And on another note I have a question about this build (and my workflow). Previously I had been processing about 20km-long flights in four 5km-long chunks, with vertices set at 50 million, to try to maximize detail. Disadvantage is that I had to enter many of the GCPs twice since I am overlapping the chunks to improve the fit.
With the new pointcloud generation step, I decided to try generating the whole model at medium (sparse->dense->mesh), inputting all of the GCPs, then regenerating the whole model at a higher resolution. I figure at worse case I can clip the model into chunks again and regenerate those if it gets too slow, but things seem to be working well. My big question is - to maintain the same grid resolution, should I increase my faces to ~150 million? that seems... daring...
-
Hello
I'm MAC user on Snow.
I have a blocking of the complete computer during blending texture sometimes.
Regards
-
clardic - try monitor CPU usage, if it blocking it is because code is efficiently written and all cpu resources are used for the blending subroutine......
-
Thanks for your fast reply. You suggest to use 3/4 CPU instead 4/4 ?
-
best if the workstation on what is processed do just reconstruction.... setting the CPU amount in settings will have no impact on this step...... just enjoy cup of coffee :-D
-
Hi Clardic
In the application preferences, OpenCL tab, do not activate any OpenCL devices. Use only the CPU cores available. I was having the same problem (OSX; Mountain Lion) and this solved my problem without any major slow down of the reconstruction process. Something must be made to avoid this overflow of the graphical card, until the machine becomes unusable. Before deactivating opencl devices, I had to force the restart of the machine to be able to use it again. This wont happen in small datasets but it's almost inevitable if you have more than 150 photos and use high quality point cloud.
Cheers
-
I have a strange thing in PS 1.0 build 1472 - the View>Perspective/Orthographic option has a "5" next to it. Just noticed after optimization and a batch dense point cloud/mesh build when I came in this AM. screenshot attached. Will update later today.
[edit] Thank you chadfx for explaining
-
It's a hotkey for toggling the Perspective/Orthographic views. It seems there are a number of non-modifier hotkeys in there if you look under the Predefined Views submenu.
-
Nice improvement on this version.
The photo matching still need improvement as photosets that I'm able to match with other softs fail to match in photoscan. Shall I increase the point number analized in the matching?
Som coments:
As seen in the atached images some tiles are visible in the point clouds, probably due to how the app processes the data, the visible tile borders is not the worst but some of the tiles having less point resolution so the point cloud has no unirom density. There is a shot where it is more visible because theh Microstation reduces the point count by a % when rotating the scene.
-
Thanks RHenriques,
The Open CL was deactivated !
I'm on 3/4 CPU core and always the same problem...
Do to investigate again
-
Hello
It's blocking also for Mersh construction.
34 pictures about 8 to 9 Mo by picture.
Mac leopart Snow
It seams that we don't have the problem with the version 9.1
Probably a problem due to the version 10
-
I noticed that I can't make models with as much detail - as many vertices in the mesh - from PS 10. I set Workflow/batch/build mesh/face count=custom, and custom face count to 50 000 000, and I was only able to get 23 669 280 faces and 11 840 381 vertices.
I am using Quality=high, depth filtering = moderate in the point cloud. I haven't really been able to resolve individual logs on gravel bars since 0.8.5. That is, in 0.8.5 the DSM more faithfully reproduced smaller logs than it does now.
There is great improvement in orthophoto generation speed in this version I think.
Andy
-
Can anyone tell me what the default settings for 'classify ground points' are (max distance, max angle, cell size)? I have adjusted mine and forgotten what the defaults were. Every time I close and re-open PS it maintains my last set parameters and does not return to the default.
Many thanks.
-
Hello clardic,
Could you please send us the log file and reconstruction parameters used?
Hello Andy,
Custom face count parameter defines number of faces the model should be decimated to if reconstruction provides bigger number. The maximal number of faces is mostly determined by the quality selected, so if you've got less faces than in face count field, this means that you've got the maximal possible faces for the parameters used.
-
Can anyone tell me what the default settings for 'classify ground points' are (max distance, max angle, cell size)? I have adjusted mine and forgotten what the defaults were. Every time I close and re-open PS it maintains my last set parameters and does not return to the default.
Many thanks.
max distance: 15
max angle: 1
cell size: 50
in build 1754
-
Can anyone tell me what the default settings for 'classify ground points' are (max distance, max angle, cell size)? I have adjusted mine and forgotten what the defaults were. Every time I close and re-open PS it maintains my last set parameters and does not return to the default.
Many thanks.
max distance: 15
max angle: 1
cell size: 50
in build 1754
Thanks!
-
Hi Alexey, thank you for clarifying.
I am having some unexpected behavior - and I wonder if it is from some behavior in the "custom" dialog for batch processing the mesh creation.
My steps - Initially processed pointcloud on medium Pointcloud settings, and medium mesh settings. Then after setting GCPs reprocessed on high and custom 50 000 000 ... (or I thought I did), but after processing I only had 5 000 000 faces. I reset processing to 50 000 000 and ran the batch and again I only have 5 000 000 faces. I went in to look at my batch parameters and it looks like they either were not updated, or were re-truncated to 5 000 000 again?!
I know that the dense pointcloud was reprocessed on high settings. I have 104 125 440 points, and there are hundreds of points per face on the big points.
Either way, my mesh once again is 5 000 000 faces. Two questions:
(1) If I specify face count "high" in mesh creation, is that the highest face count I can get for a give dense point cloud?
(2) If I specify a custom face count that is higher than the end result at "high", is it possible that the software "adjusts" my custom face count number (by removing a zero) during/before processing?
otherwise I mistakenly entered 5 000 000 instead of 50 000 000 twice, and since I always count my zeros when I enter them in the millions or more, I am finding that hard to believe.
Reprocessing on "high" now.
[EDIT] I just noticed that if I go into the batch window and <edit> the batch parameters, then cancel out of the batch processing after editing that it doesn't save the edits. Maybe I canceled to check something like a point count before going back in and didn't check before starting mesh generation. So this might be user error.
-
Hello Andy,
You can always check reconstruction parameters used in the Chunk Info from the Workspace pane.
As for the High, Medium and Low presets in Build Mesh dialog they are based on the number of points in the Dense cloud. But if you use Custom option you should get either this value or the maximal possible for the current quality of dense cloud.
-
I am definitely struggling with the 'Classify Ground Points'.
Initial run on default settings seemed to work ok.
I then ran it again with very different settings and very little seemed to happen.
I have since tried numerous other settings and again, very little (if anything), is happening.
I am leaving the parameter 'any class' selected in the 'from' box and have also tried a few of the others.
Is this process not yet fully functional or am I not doing something right?? I am using build 1745
It seems like it will be a very impressive to function to have and I'd love to be able to figure it out!! It would also be good to be able to export a point file (las etc) of individual class types.
-
I am struggling with the mesh quality still. I generated point clouds on "high" setting and have a detailed surface with ~100 000 000 points, but when I generate a mesh on "high" setting I only get 18 000 000 faces, (would prefer more like 50 000 000 to keep the detail) and the surface appears triangulated and much more artificial than the pointcloud surface.
If I generate the pointcloud with less smoothing, will that result in a mesh that has more faces?
[Edit] - I also noticed that in the "info" option for the mesh, there is a depth filtering parameter, but no ability to change it in the batch - is that just bookkeeping that's keeping track of the pointcloud generation parameters, or is there a "moderate" depth filter being applied twice?
-
Hello airmap3d,
That's because Classification to Ground assigns Ground class to points matching criteria in the cloud but does not take off already assigned classes if the point doesn't match the classification criterion.
So if you wish to perform ground classification with different parameters you need to assign any different class to ground class points: select all the points in the dense cloud, select Assign Class from Tools menu, select From Ground to any other (unclassified, for example). Then just start classification.
Hello Andy,
Depth filtering parameter in the chunks info refers to the depth maps/dense cloud generation it could be duplicated twice if both model and dense cloud are present in the chunk.
Polygon count presets (High, Medium and Low) in Build Mesh dialog are based on the number of points in the dense cloud, if you require more faces than highest preset than you should use Custom input. The filtering parameter doesn't matter for the numbers automatically calculated in the polygon count field.
-
Got it. Thanks Alexey.
-
I was having some issues with my scripts, even after adapting for the new api parameters. I think I finally traced it to a problem with the buildModel function when I assign an arbitrary number of faces rather than a 'low', 'medium', or 'high' setting.
i.e.:
import PhotoScan
doc = PhotoScan.app.document
chunk = doc.activeChunk
chunk.buildModel(surface='arbitrary', source='sparse', interpolation='enabled', faces=100000)
The same problem seems to apply across projects and regardless of 'sparse' vs 'dense'.
Here's the error I'm getting:
TypeError: bad argument type for built-in operation
I won't give the whole traceback, since I think it is not working properly. I was scratching my head chasing several lines of code it was referring to in tracebacks. After investigation, all the lines I was tracing were just fine. I think it was just referencing them incorrectly.
Of course, there's a chance I'm just doing something wrong...
-
I have an odd floating divider in my workspace pane. see attached images. v 1.0 build 1748. The first image shows the pane with the divider in the middle (left) and bottom (right) of the workspace pane. When it's on the bottom docked next to my console pane, there's an extra gray space when comparing the two (second image).
Andy
-
Andy, if you select a photo from the chunk in the workspace pane it will be displayed in the pane below under that divider. If the divider isn't visible but you have the extra gray space then you can drag that up to make the photo preview/info pane visible.
-
Thanks James. Never noticed that before. Guess I don't need it... ;D
-
Hello captured,
Thank you for reporting. The problem will be fixed in the next update.
-
I don't know if this is me or PS 1.0, but in 1.0 when placing GCP, I have to re-apply the filter-by-marker option more times, and unplaced markers are further away from the actual targets on the ground before I place a few (after establishing initial ground control plane). I use the quick load method to load GCPs in from previous surveys of the same area, then I place them.
In 0.9.x, after the first three points were entered to define a plane, I could just update as I placed imported GCPs and when I filtered by GCP I would typically get all but one or two images when working my way up the flight path and regularly updating ground control. Then I could go back through the project when I'd entered GCPs and look for any unplaced markers.
Now when I am entering GCPs, I have to re-apply the "filter-by-marker" when I am finished and I will get 2-3x as many images as initially, which I then have to place GCP on.
One complicating factor that could make it my fault is that in previous surveys I loaded the imagery in ~200 image ~5 km chunks, and I am doing this one as a single 30 km project to eliminate duplicate entry of GCPs (1628 cameras, 198 GCPs). But I don't know the details of how the marker filter feature works, so I am not sure if it should be expected to affect image selection behavior like it does, given a longer ribbon type aerial survey.
-
I also just noticed that when editing the sparse point cloud if I have part of my model disappeared in the foreground and I select points, the disappeared points within range of my selection get selected too even though they are not visible.
-
Tried to regenerate dense point cloud and mesh after entering GCPs and got interesting results. I generated the initial dense point cloud and mesh on "low" setting, and tried this one on "ultra".
I imagine I'll have to go back to my 256GB RAM machine or try it on a lower setting, but I was surprised my CPU didn't do any work according to the log. I am posting more because of the lack of CPU numbers than the "bad allocation" error. Guessing that means I should have tried a lower resolution reconstruction.
Project is 1 chunk, 1628 12MP images, 119 markers.
relevant part of log follows:Using device: GeForce GTX 670, 7 compute units, 2048 MB global memory
max work group size 1024
max work item sizes [1024, 1024, 64]
max mem alloc size 512 MB
initializing...
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: unable to estimate disparity range
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
skipping pair: unable to estimate disparity range
skipping pair: unable to estimate disparity range
skipping pair: unable to estimate disparity range
skipping pair: unable to estimate disparity range
skipping pair: camera visibility test
selected 1501 cameras from 1519 in 490.592 sec
Loading photos...
Reconstructing depth...
[GPU] estimating 1063x3768x320 disparity using 532x942x8u tiles, offset 0
timings: rectify: 18.408 disparity: 77.901 borders: 0.881 filter: 3.188 fill: 0
[GPU] estimating 1736x3920x384 disparity using 868x784x8u tiles, offset 0
timings: rectify: 0.356 disparity: 11.373 borders: 0.097 filter: 1.955 fill: 0
[GPU] estimating 2524x4042x256 disparity using 842x809x8u tiles, offset 0
[CPU] estimating 4398x3994x224 disparity using 1100x999x8u tiles, offset 0
Error: bad allocation
timings: rectify: 0.306 disparity: 13.552 borders: 0.138 filter: 8.249 fill: 0
[GPU] estimating 2432x3989x256 disparity using 811x798x8u tiles, offset 0
timings: rectify: 0.37 disparity: 7.254 borders: 0.143 filter: 2.24 fill: 0
[GPU] estimating 1803x3967x416 disparity using 902x794x8u tiles, offset 0
timings: rectify: 0.24 disparity: 6.969 borders: 0.101 filter: 1.133 fill: 0
[GPU] estimating 3332x4806x352 disparity using 833x801x8u tiles, offset 0
timings: rectify: 1.056 disparity: 14.842 borders: 0.101 filter: 1.207 fill: 0
[GPU] estimating 4579x4532x288 disparity using 764x907x8u tiles, offset 0
ocl_engine.cpp line 298: clEnqueueNDRangeKernel failed, CL_MEM_OBJECT_ALLOCATION_FAILURE
GPU processing failed, switching to CPU mode
[CPU] estimating 4579x4532x288 disparity using 1145x1133x8u tiles, offset 0
Error: bad allocation
finished depth reconstruction in 202.769 seconds
Device 1 performance: 0 million samples/sec (CPU)
Device 2 performance: 197.047 million samples/sec (GeForce GTX 670)
Total performance: 197.047 million samples/sec
Error: bad allocation
Finished processing in 1813.53 sec (exit code -1)
-
Hello Andy,
Could you please check that 64-bit version of PhotoScan is installed. "Bad allocation" problem may be connected with 32-bit version.
As for the Photos pane filtering, filter is applied only when you select any "Filter by" option, So when you change current marker projections sub-set of photos in Photos pane remain unchanged and will be updates only after "Filter by" command is selected again.
It is possible to filer photos by several markers - in this case "gray flag" icon next to photo thumbnail in the Photos pane will indicate that there is at least one unplaced projection for selected markers on the current photo.
-
64-bit version is installed. It's possible I went over memory. This machine only has 48GB. Was surprised at the CPU not working at all though. I re-started a batch with dense cloud on "high" setting and successfully generated the dense pointcloud with CPU and GPU both working.
I've been stuck at 87% on the "generating mesh..." step now for about 4 hours with only 5% CPU working. Should I be using more CPU?
Most recent message in console window is:
constructed triangulation from 14322214 vertices, 28644422 faces
grid interpolated in 21635.5 sec
triangulating...
I've attached a screenshot showing that it looks like PS is using one CPU at a time and doing some kind of "hopping" (I only have twelve cores, but hyperthreading is enabled in BIOS)
-
there are steps when just one core is running, its absolutely normal, as not all parts can be multi threaded. the process just jumps form core to core........
-
Just wanted to let you know of a possible bug in the latest Beta build. When I selected camera correction in the Ground Control settings, my model would align photos just fine, sometimes at 4 hours in length. Just before completion, Photoscan crashed and I had to start over. When I removed camera correction from ground control, it went back to working just fine; however, the models I am building are still being built upside down and I believe utilizing the camera correction would solve this instead of the extra step of manually doing it myself.
I can test it again and try to record the logs, if you would like.
-
Hello Michael,
Could you please check if the crashes still occurs with the latest build installed (1748)?
Please note that camera correction option works only if there are camera coordinates input in the Ground Control pane. Please describe the problem with the coordintate system - do you have any camera coordinates? If not than you need to use Rotate Object tool.
-
Hello guys,
is it still possible to set up a reference distance in a scene with Pscan?
example attached
GR
-
Hello George,
To create a scale-bar in the version 1.0.0 you still need a pair of markers that should be selected in Workspace or Ground control pane, or in model view, then you need to select "create scale bar" option from the context menu available by right-clicking.
-
would that be enough to make the model scaled i.e. to be able to obtain dimensional features from it?
-
one more question ...
is it possible to reorient model but not in a manual way?
-
Hello George,
If the model is scaled using reference distance you can measure the relative distances between any points on it.
Model can be oriented using:
- "rotate object" tool,
- ground control pane data,
- Python scripting.
-
Creating dense point cloud with 221 photos, settings ultra high and mild, is taking days to process. Reconstructing depth took a little over 24 hours. Generating dense point cloud has been running for about 11.5 hours and indicates 49.5 hours left.
System:
Intel Core i7-3370k @ 3.50GHz
32.0 GB ram
GTX 660 3GB ram
GTX 460 1GB ram
-
Hello bisenberger,
And what is the current memory usage?
-
Hello Michael,
Could you please check if the crashes still occurs with the latest build installed (1748)?
Please note that camera correction option works only if there are camera coordinates input in the Ground Control pane. Please describe the problem with the coordintate system - do you have any camera coordinates? If not than you need to use Rotate Object tool.
Hi Alexey,
Sorry for the late reply. I will check and let you know.
As far as coordinates, we only have coordinates for the markers initially, not the cameras.
Best,
Michael
-
Hello bisenberger,
And what is the current memory usage?
29.9/31.9 GB (94%)
-
Is there a possibility to acquire official fully functioning AP1.0.0 b. not only demo version which does not include export possibilities and so I cannot test them ?
-
Hello janoz,
If you already have license key, you can use it for PhotoScan 1.0.0 activation. Otherwise you can request PhotoScan Pro 3-day trial license key from our web-site: http://www.agisoft.ru/products/photoscan/professional/trial/
In case you are interested in Standard edition Trial please contact our support team.
-
I noticed that if I select a portion of the model, then invert the selection, the model selection is inverted, but the camera selection is not. I don't know if this occurred in earlier versions, but it prevents me from quickly cropping a large project into separate chunks after GCPs are entered.
-
Hello Andy,
Delete, Crop and Invert selection tools only work for points of cloud and faces of mesh.
-
I still can't make DSMs with as much detail as I could in 0.9.1 and 0.8.5. Even when I use "custom" settings for faces, I can't get as high as the 50 000 000 faces I was able to get before. Max is about 30 000 000 when I choose custom, and the face density is way less dense than the point density.
I attached some images trying to show the loss of detail I am seeing. Before I could export a 0.5 m DSM and often see individual logs and subtle drainage features. Now I am lucky to see either, and the surface is much more faceted.
Processing settings were medium for align, high for pointcloud, and custom/50 000 000 for mesh.
on a related note, it would be great if I could produce a DSM directly from the pointcloud by binning the elevation values per grid cell (median, min, max, and mean would all be good options). Then I wouldn't even need to generate the mesh at a high resolution for the orthophoto.
One other note, although this has been the case in 0.9x and 0.85 too - when I have pre-imported GCPs, and I stepp my way through the GCPs filtering by marker and pinning GCPs, even if I press the "update" icon to update alignment, I don't see new photos appear in the "Photos" window until I "filter by marker" for another marker, then go back to the one I was working on.
-
I am having the same thing happen as in an earlier post (http://www.agisoft.ru/forum/index.php?topic=1476.msg8441#msg8441), no progress from 87% done in "generating mesh" task for several hours now, machine is only using ~10% of my CPU, and looks like a lot of CPU-hopping going on between differenct cores in my first CPU (dual CPU Xeon machine).
Only difference this time is that I know I didn't go over memory - this machine shows only 28GB of 192GB is being used. Screenshot from task manager attached.
This is on build 1748
-
Hello Andy,
At 87% of mesh generation phase mesh is being triangulated at this step really doesn't use all the cores. It may take quite a long time depending on the dense point cloud quality (and number of points in the cloud).
-
Hi Alexey,
Thanks for the clarification.
-
hi all
what is new in build 1750? any changes to ortho rendering?
GR
-
Hello George,
Bug fixes, French GUI (not yet complete) and extensions in Python functionality (like meta data for Frame, Camera, Marker and Scalebar classes and access to image EXIF and image quality data).
-
Is the pre-release stable on Mac OSX 10.9? I am thinking of upgrading, especially since it has the multiple screen capability.
-
Buying a PhotoScan Pro license and starting up production, would you recommend starting out with the 1.0.0 pre release right away, or better begin with the "old" released version until official release?
Is there an estimate date for official release of the 1.0.0 (or best guess)?
Tom
-
I always use the latest beta versions for production and never had serious problems so I really can recommend to use 1.0.0 from the start. And the actual version again seems to be remarkably stable.
-
I notice on v1.0 that there's still no error trapping for jpg with one (or both) dimension(s) > 65535. The JPG standard only supports up to 65535x65535, and JPG export fails if it's larger than that, but PS goes through the whole selection and blending process before saying "can't export ortho", rather than stopping after it calculates max dimensions > 65535. I don't know that it matters much, but a message saying "JPG too big" rather than waiting 10-30 minutes and then getting "can't export jpg" would at least explain to users what's happening, and would save a lot of time. As of now you have to watch for the size message and manually cancel.
Andy
-
Having some trouble with Verold 3d Viewer, when i upload the model i get x- axis where i should have my Z-axis wich messes up everyting on verold.
I have tried to rotate object in agisoft before uploading to switch axis but it dosent help, maybe somewhere in the steps i have it wrong, has anyone tried this?
On sketch fab there is the possibility to swithc axis in the browser on their site so no problem there.
-
Unless you have the Pro version of PhotoScan, you cannot re-orient your object in the software. You should be able to do it in a 3rd party application before uploading to Verold, though. It actually is nicer for Sketchfab as well, so your object doesn't rotate with odd stop points in the rotation axes, as happens when you try to adjust the axes on their website.
Cheers, -C
-
i have pro version
-
Hi Peter, perhaps this helpful post created by Lee will be of use, then?
http://ir-ltd.net/delivering-aligned-and-scaled-photoscan-outputs/
Cheers, -C
-
Hi Alexey,
I am having issues with the classify point cloud functionality. Runs very quickly in 1736 but no luck with build 1760. In 1736 it just initialized for a few seconds and then away it went now it just stalls after the second initialize. I should add that I have been giving up after fifteen minutes based on previous experience and reverting to build 1736.
Also has the softness/blur issue in the ortho export been resolved in the 1760 build as I have had to revert to 9.1 for production ortho export my forested areas or areas of very complex geometry.
Thanks again to the team for the awesome product.
-
Hi Everybody...
Some new features to implement in this new version:
- Add 6 usefull view buttons (Upper view, bottom, front, back, left and right)
- Add some usefull combo for navigation mode (for example ctrl + left click to rotate pointcloud or model around z-axis, shift + left click to rotate pointcloud/3D model around x-axis...)
- Add a mesh edit mode (create polylines to bound mesh on street or clean surface, delete manually spikes or wrong vertex...)
- Add automatic identification of planes in meshed model for geotechnical purposes...
These are only suggestions.. ;)
Thanks agisoft team!
-
Matt,
In 1760 I am getting clean orthos (0.15 m ortho and 0.5m dem from canon d12 at 500-750 meters elevation) where the detail looks as good as I saw in 0.8.5 and better than 0.9.1 I think. Settings are high alignment, ultra high density point cloud, and custom 50 000 000 face mesh. I think it's slower than 0.9.1 though, and the water noise might actually be worse - I tried it on moderate, which is supposedly the default values from 0.8.5?
I miss having more control over noise reduction.
Also I have had a crash while batch-building mesh (I think during a save) that corrupted a project file, so beware.
And when I try to view the mesh I get a timeout error on my video card (but I am running a mixed AMD/Invidia setup). Dense point cloud viewing/editing works fine. I wish there was a way to remove disconnected clusters of points from the main pointcloud because those are mostly blunders.
-
Hello Peter,
In the version 1.0.0 there's "Rotate object" tool on the toolbar that can be used to reorient model relatively to the coordinate system axis prior to the export or model upload.
Hello Matt,
Have you performed the classification on the same project with the same number of points in the dense cloud?
-
Hello piste,
1. Add 6 usefull view buttons (Upper view, bottom, front, back, left and right)
2. Add some usefull combo for navigation mode (for example ctrl + left click to rotate pointcloud or model around z-axis, shift + left click to rotate pointcloud/3D model around x-axis...)
1. This is already implemented in 1.0.0 - you can use NumPad hotkeys or corresponding view options in the view menu.
2. For such rotation purposes you can use red, green and blue circles located on the trackball, thus rotation will be performed only around the corresponding axis and other rotations will be locked.
-
Hi Andyroo,
Thanks for the input. The main area I am having trouble with in the orthos is in dense bush where I get this 'fog' below the canopy. I can work around this by classifying the dense cloud and filtering out everything but the ground (build 1736) then exporting the project back to 9.1 and exporting the mosaic from there (its not ideal).
I think we are doing much the same thing in the way of time series work so it may be worth having a chat sometime. Although I suspect I am on the opposite side of the earth to you on the land down under. I have been sucessfully running the AMD/Nvidia combo for over a year now with no real issues. I have stopped upgrading the nvidia drivers at 311.06 as I am running a GTX 590 and run the most up to date catalyst drivers. No Stalling when I am viewing meshes although I never really bother with anything over 40 mill on my processing machine. I suspect the Titan 6 GB would solve your issues with large mesh display but it would still be pretty bad in the openCL processing side.
HI Alexey,
Yep tried it with numerous dense clouds at everything from low through to ultra high, they all work in 1736 not at all in 1760. Given up using 1760 at this stage as I am under the pump with projects. I will try and have another look at it tonight while I am processing.
Thanks
-
Version 1.0 1760 has issues and is very unstable under MacOS 10.9 Mavericks.
Main issues are:
- Some camera position polygons are black instead of blue.
- Box limits are strange pixelated black lines, instead of smooth gray ones.
- If the "Rotate Region tool" is selected, the box becomes black. It's impossible to see anything inside it (see image).
- While select Batch and try to change parameters in any process, PS crashes immediately.
There are some annoyances in the interface, mainly related with displacement of feature access. Menus have disappeared (for instance "view matches") and now have a more awkward way to get it (second click in the photos pane). The use of an image scale is now less obvious than in previous versions. As I already said in a previous reply, about options for photo filtering, any software should always provide, at least, 2 ways to get to a feature among a menu command, a button or a right mouse click pop up menu. PS, lately, tends to abuse of the latest option without provide any other. Functions are there indeed but are less obvious for most users.
Keep on the good work.
-
HI Alexey,
Still cannot get the classify point cloud to work on build 1760. Basically freezes the software and needs to be shut down via task manager.
-
Version 1.0.0 build 1760.
Build Dense Cloud with 166 cameras, settings Ultra high/Mild, running for almost 2.5 hours showing 0% done, CPU at 15%, and memory at 9%. Tried to cancel process, but wouldn't, so had to end process with task manager.
It also seems like photo matching takes longer than in version 0.9.1.
-
Hello Matt,
We'll check that.
Hello RHenriques,
We also see some visualization issues on Mavericks, while on the previous MacOS PhotoScan works fine. However, Batch Process crashes under Mavericks will be fixed in the next PhotoScan update.
As for the GUI suggestions we are trying to make the interface easier, for example, View Matches in 1.0.0 allows to click on exact photo in the Photos pane and see its matches with the rest of images, whereas in previous versions one had to search for the photo label in the long list. But still we are considering different options and if you have any additional suggestions we will appreciate them.
Hello bisenberger,
What are the computer specifications and memory commit in Task Manager?
-
Intel Core i7-3770K CPU @ 3.50GHz
32.0 GB RAM
GTX 660 3 GB
GTX 460 1 GB
Memory 2.6/31.9 GB (8%) it was at 9% when running Build Dense Cloud mentioned in previous post.
-
When using Gradual Selection - Reprojection error on sparce point cloud it selects a number of points. If I delete the points and run Gradual Selection - Reprojection error again, it keeps selecting about the same number of points.
-
Hello bisenberger,
Have you got the log file from that processing? I've just finished Ultra reconstruction for 150 image project and got memory consumption peak about 26 GB. And the processing time was about 17 hours for depth maps reconstruction (double-Xeon workstatation with single GTX580) and 1.5 hours for dense point cloud generation.
-
Hi Alexey,
Here is a link to my log file:
http://www.digital-mapping.net/forums/Photoscan/2013/November/log_pro.zip
Does the log file keep the history for all projects? I have ran a few projects since the problem one. If not let me know and I'll run the project with the same settings again and resend.
When you ran your test, did you have Depth filtering set at mild for Build Dense Cloud?
-
If build a dense point cloud in a project without ground control points and later add ground control points, does Build Dense Cloud... need to be ran again or is the existing dense point cloud adjusted?
-
you can add gcps to georeference the model and the dense cloud will be preserved, as this does not affect relative camera positions or internal camera parameters.
However, if you then use the gcps to optimise the solution then relative positions and camera parameters will be modified, and the dense cloud etc will be need to be regenerated.
-
Hello bisenberger,
On my tests I've used Aggressive filterring, but I think I can restart it with Mild as well.
Thank you for Log file, but actually it contain many processes, so if you can point to the right one it will easier the task.
-
The 64-bit windows download link to the 1.0 pre-release seems to be broken...
-
Hello aphextwin,
We are currently updating the installation packages, so please wait a little bit until new build will be available.
-
Generating a texture with color correction seems to be frozen. Can't cancel the process. I've run the same routine on a few other projects and didn't encounter this problem. CPU 100% memory 13%.
log file
http://www.digital-mapping.net/forums/Photoscan/2013/November/log_color_correction_freeze.zip (http://www.digital-mapping.net/forums/Photoscan/2013/November/log_color_correction_freeze.zip)
This is similar to what happened with the Build Dense Cloud issue I mentioned in an earlier post.
-
Hello bisenberger,
Color correction works slow compared to the unchecked option, so according to the log file (unless to don't see any changes in the console pane) the operation is in progress.
-
Hi Alexey and Dmitry
Some of the graphical issues, pointed in my latest reply, still remain in this latest 1768 version. The worst one is the render region black box when the Rotate region tool is selected. The crashes while selecting parameters in the batch option are gone. All options are working just fine now.
Cheers
-
For the last view versions (1760 and 1768 I think) I haven't been able to view the mesh without getting error code 3 (560 Ti - timeout) or error code 6 (670 - hardware config doesn't meet minimum specs for the application) messages on my NVidia cards ( 2 different machines - (GTX 670 in one and GTX 560Ti + HD7970 in the other). It might have been occurring longer, but I usually don't open in mesh view.
If I open the file right after I start PS, it goes to mesh view and crashes. If I click on sparse (or dense) mesh button before opening the project file it will stay in pointcloud mode. It would be nice if it defaulted to sparse cloud when you first open a project, since it's faster to load and switch chunks.
My latest project is 49 921 267 faces, 24 975 489 vertices (Isolated to 1 chunk, saved in 0.8.5 format or latest).
-
Andy you hit the wall of GPU ram. for cca 50 mil trias you need cca 3GB VRAM on GPU......
-
I've got a file open right now in 1768 on my MacBookPro (GT650M 1Gb) that is 50.7 mil faces. No crashes. Interactivity is a blazing 1.3fps, though. ;-)
Is it just me, or is MacOS a bit less fussy with memory requirements than you guys run into on the PC side? I've definitely run into maxed out situations where I'll start getting into a VM swap slowdown, but it will eventually finish most of the time and doesn't completely lock up the machine.
-
Chadfx might be right. Wishgranter requirements come with a bit of surprise for me because I own a MacbookPro (3.06GHz core 2 duo) with a old NVIDIA GeForce 9600M GT 512 MB and it behaves just fine with 50 million faces models. The interaction with the model is not very fast but renders correctly and it's usable. Things get a lot slower with models above 80 million thought. Even so, they render just fine.
It's strange because usually OpenCL drives versions lag a bit in MacOSX when compared with Windows. Even so, for most graphical tasks, I've always liked MacOSX better.
Cheers
-
Yeah, I think you and I are the lone Mac users around these parts. ;-) But overall the software hasn't given me too much trouble and it doesn't seem to be as much of a memory and resource hog as on the Windows side of things.
I wonder if you run into the same issue I get every once in a while, when all of the menus become grayed out (usually after some processing step). I can get everything to come back if I open up the preferences via the hotkey shortcut (command - ,). I haven't found a reproducible case to submit, but it seems to be in every version I've used, including these pre-releases).
-
you can add gcps to georeference the model and the dense cloud will be preserved, as this does not affect relative camera positions or internal camera parameters.
However, if you then use the gcps to optimise the solution then relative positions and camera parameters will be modified, and the dense cloud etc will be need to be regenerated.
Thanks James
-
I was creating a dense point cloud with ultra high/aggressive settings. It was almost completed when I got this "Not enough memory" message.
(http://www.digital-mapping.net/forums/Photoscan/2013/November/not_enough_memory.jpg)
-
Hi Bisenberger,
You need more RAM. Ultra High reconstruction uses lots of memory try it on High. Ram is cheap, kits of 64 GB Kingston ECC RAM can be found for as low as $800.
Hi Alexey,
I have tested build 1768 and the point classification feature is still not working. It looks as though you have implemented a new option but neither of these options get past the second initialization step. Build 1736 classifies all datasets I try.
Thanks
-
Hi Alexey and Dmitry
Some of the graphical issues, pointed in my latest reply, still remain in this latest 1768 version. The worst one is the render region black box when the Rotate region tool is selected.
Hello RHenriques,
We are working of this issue.
-
Hello bisenberger,
I'll try to reproduce this problem on the similar dataset or at least check the memory peak during the processing. I see that there are 194 cameras and what was the resolution?
Hello Matt,
Can you send us the log file created during classification? We've tried to reproduce the problem, but the classification worked on our samples. Probably you can provide some additional info like coordinate system, number of points in cloud, parameters used?
-
Hello bisenberger,
I'll try to reproduce this problem on the similar dataset or at least check the memory peak during the processing. I see that there are 194 cameras and what was the resolution?
3264 x 2448 (Canon A590 8.0 MEGA PIXELS)
-
Hello bisenberger,
Currently I'm in the middle of processing 200 images with 21 MPix and memory peak is about 32 GB so far. But it also depend on the overlap ratio. Could you please post a screenshot of the camera positions?
-
Hello bisenberger,
Currently I'm in the middle of processing 200 images with 21 MPix and memory peak is about 32 GB so far. But it also depend on the overlap ratio. Could you please post a screenshot of the camera positions?
KAP project:
(http://www.digital-mapping.net/forums/Photoscan/2013/November/camera_locatons.jpg)
Alexey, thanks for clarifying the photo overlap question :)
-
Hello bisenberger,
Very high overlap can result in higher memory consumption, so think that disabling photos in the area of super-overlap will help to get under the memory limit.
-
Hello bisenberger,
Very high overlap can result in higher memory consumption, so think that disabling photos in the area of super-overlap will help to get under the memory limit.
Thanks Alexey, after I add ground control points I'll weed out the lower altitude photos and run it again with the photos closer to the same altitude.
-
Chadfx
A while ago I told Agisoft Team about the greyed menus. I also got that issue each time I finish any intensive process. I was able to avoid that exactly using the same technique you are using :-)
Recently I've found that you only need to use the Photoscan Menu -> About PhotoscanPro. You will get the about dialogue, press ok and you'll get the menus back. This menu does not suffer from the grey syndrome! Since it's a low annoying issue, I did not bother the Agisoft team anymore about it. They must be concentrated in most important software features and I'll be happy with that ;).
This package has been super stable under MacOSX. There was an issue with the batch processing in the last 1760 version but it was quickly corrected.
Cheers
-
Oh yeah, I guess that would work, too. Glad it wasn't just me at least! I agree it's not too big of a deal, and I can't always get it to happen, so it's not an easy bug to track down.
The MacOS version has definitely been very stable...better than a lot of the other tools out there!
-
Win64 Standard reports a error in the products.cab while installing. Extracting with 7zip does not work, too. There might be an error in the msi file?
-
Hello Ricky,
I've just downloaded Win64 Standard package for PhotoScan 1.0.0 and it has installed correctly, please try to download it once again.
-
Serial number
Hallo, will my serial number for version 0.9 works for beta version 1.0?
Thanks.
-
Hello Robert,
Yes, your license key is valid for all versions up to 1.9.x (but you should make sure that you've downloaded correct edition of PhotoScan).
-
Hello Ricky,
I've just downloaded Win64 Standard package for PhotoScan 1.0.0 and it has installed correctly, please try to download it once again.
Now it works
Thx!
Ricky
-
Hello,
Has merging of chunks changed? in 0.9.1 I usually build chunks (about 16) separately and then merge into one mesh. Most of the time i got better details then generating mesh in one pass. In 1.0.0 build 1768 standard, result is just joined chunks, meshes from chunks are not connected.. Or did I miss something?
Thanks.
c.
-
Hello carbon,
In the version 0.9.1 there was "merge models" option which produced low quality model based on the original chunks models, in the 1.0.0 it was changed to "combine models" that allows to merge high quality meshes unchanged.
-
And is there any way to get "flattened" mesh from chunk meshes?
Thanks for reply.
c.
-
Hello carbon,
You can create sparse point cloud based geometry on the Merged chunk. So this option doesn't require to build dense clouds and models in the original chunks.
-
Hmm, I don't think this is what I need. Maybe I have to better explain my pipeline.
1 load all photos
2 align them
3 resize chunk region to bound smaller part of object
4 generate mesh at high or ultra high setting at 2 mil. faces
5 duplicate chunk move region to next part with overlap --> 4
loop until all parts are generated
6 align chunks by cameras
7 merge chunks and merge meshes
8 decimate resulting mesh
this way i get better results then with just single pass generation with region around whole object
so, i can do this in 1.0.0 because meshes are not merged in step 7, would it be possible to have that option back?
thanks
c.
-
Hello carbon,
Merging models in 0.9.1 and earlier versions produces low-quality resulting model, irrespectively to the quality used in the original chunks - so it is unreasonable to use High/Ultra quality.
I suppose that you will be able to reconstruct mesh for the whole original chunk using at least Low quality and I don't think it will be worse than the merge models result.
As an alternative, I can suggest to use Python scripting to split original chunk bounding box into regular grid of adjunct smaller size regions for duplicated chunks. Then you'll be able to process each one in High or Ultra and combine the models (in the version 1.0.0) without any loss of details.
And another comment regarding the step 6. If you reference the chunk prior to duplicating you can skip this step, since all the duplicated chunks will be referenced in the same coordinate system.
-
Some of the graphical issues, pointed in my latest reply, still remain in this latest 1768 version. The worst one is the render region black box when the Rotate region tool is selected.
Please check the latest build 1771, visualization issues should be fixed now for Mavericks.
-
Hi Alexey
Confirmed!! All graphical issues I had detected in Mavericks are now gone. I don't know if you did any other optimizations, and can only be just a "feeling", but this version seems to be faster than the previous ones while rotating models with texture or solid mesh view mode.
Cheers
-
Build 1768 - I noticed that orthophoto export doesn't honor the bounding box extents if the box is changed after mesh is generated from sparse point cloud (to select a subset of the model for ortho export).
-
Hello Andy,
Orthophoto is generated for the entire 3D model, unless Region section is used in the Export Orthophoto dialog.
-
Hi Alexey,
I am still seeing graphical issues with the 1771 version and MacOS Snow Leopard 10.8.5. The build texture process just renders a completely black texture. It was also this way in build 1768, but was fine in 1760 (and earlier versions).
I'm happy to provide more detailed info if you need it, just let me know.
Thanks! -Chad
-
Hello Chad,
And what is the texture resolution selected? Please also provide the copy of the Console information on PhotoScan start-up.
-
Hi Alexey,
I am getting black texture output at all resolutions (2048, 4096, 8192, etc). Color correction on or off. Mosaic, average, etc. I haven't tested every combination, but they all seem to give the same result. I am not seeing any errors in the console, and this is broken for every setup I have loaded. Testing a bit more, it seems to work if I set it to Min or Max Intensity, just not Average or Mosaic. All of the images I use are JPG images converted from RAW in Adobe Bridge.
Here's the console info:
OpenGL Vendor: NVIDIA Corporation
OpenGL Renderer: NVIDIA GeForce GT 650M OpenGL Engine
OpenGL Version: 2.1 NVIDIA-8.16.74 310.40.00.10f02
Maximum Texture Size: 16384
Quad Buffered Stereo: not enabled
ARB_vertex_buffer_object: supported
ARB_texture_non_power_of_two: supported
This is on a 2012 15" MacBook Pro Retina, but I am also seeing this problem on an early 2012 iMac (and can test on a couple of other machines if you are not seeing this problem on your end).
Hopefully this will help you find the issue. Please let me know if there is anything else I can provide to you.
Good luck! -Chad
-
Hello Chad,
Are you using latest build? Could you please also specify if the problem appears irrespectively to the Mac OS version (or if it is only Mavericks).
-
Hello Alexey,
I have verified that the issue is happening in the 1768 and 1771 builds in Mac OS 10.8.5 (Snow Leopard). I have not upgraded to 10.9 Mavericks yet. If I roll back to build 1760, the texture building works as expected...so I think it is a recent change that has caused this to come up.
I don't think I have changed anything in my system config, but I will try to see if I can find anything else that might be helpful.
Cheers, -C
-
Just a quick update, Alexey. I also tried this on an early 2013 iMac (10.8.5) and it had the same issue. I also tried disabling all extensions (Safe Boot mode) and the problem was still there. I also tried it in 10.8.3 and it was the same (works in 1760, broken in 1771).
Once again, it's only with Mosaic and Average texture builds, it still works if I use Max or Min intensity.
Please let me know if there is anything else you would like me to try, or even if you would like me to run a special debug version, etc.
Cheers, -C
-
Hello Chad,
Thank you for this information. We'll try to fix that as soon as possible.
-
Hey folks,
i got an error loading a project file, after switching from a previous 1.0.0 beta to the current build 1771. My file PSZ is 12GB bit containing ca. 30 chunks. Before it works without problems, so it shouldnt be a problem of RAM. After the half is loaded i get the message "Cant open file".
Whats wrong, where are logs? A clear error message or an error code with a reference to documentation would be great to make this product even more professional.
Take care,
Markus
-
Hello Chad,
Please check the latest update and report if there are still any problems.
Hello Markus,
Is it possible that the project is corrupted?
-
Hello Alexey...that fixed it!
I just did a quick test and it worked as expected. I can test on a couple of other Mac platforms later, but I think you have squashed that bug.
Thank you!
-Chad
-
Hi,
I exported an orthopho from a model and got a strange result. Seems like the texture given by some cameras is not calculated correctly. See image attached.
-
Hello Ronald,
Is it an export made from build 1774?
Just to make sure - there are no blue images in the set, correct?
-
Hi Alexey,
the export is from build 1771.
No, there were no blue images. The textured model itself (generic, mosaic 16384x1 texture size, colour correction enabled) looks great! However, if I export the orthophote in default settings (mosaic) I get this strange result. So far I was not able to reproduce it since orthophoto export took me 12 hours (around 1200 photos used).
Also, the orthophoto partly shows textures that are on the back of the building. Will try to disable all cameras I don`t need for texture building, build texture again and then try to export orthophoto with diabled cameras from backside.
-
Hello Ronald,
Could you please try build 1774, we have fixed some issues for orthophoto export there that may have appeared for complex objects (at the second issue should be fixed).
-
Major Texturing Problem
Build Texture creates either black or very pixelated textures. I'm using Build 1774, Standard Edition, 64 bit, Windows 8.1.
Attached are the screen shots of the textured meshes for copies of the same chunk, 12,314,183 points in the Dense Point Cloud, meshes with 10 million and 2 million faces. Also attached is a screen capture of the 2 million face shaded mesh.
Texturing was done using:
Blending mode Mosaic
Texture size 2048
Texture count 1
Color correction No
-
I just tried decimating the mesh to 200,000 faces and building texture. The result is all black.
I have two similar chunks with 20 million and 2 million face meshes of each. The texture results for those also range from small "pixels" to large "pixels" to black.
-
Hello David,
Thank you for reporting.
This problem may be related to the huge number of flying polygons around the model - looks like the Interpolation was disabled (if you open View Mesh UVs dialog you'll probably see a lot of small pieces). Could you please remove these parts using Connected Components criterion in the Gradual Selection dialog and rebuild the texture.
-
Attached is what I see with the View Mesh UVs dialog box for the 2 million face mesh. Same thing with the other meshes.
I used Gradual Edit at 98% Connected Components with the 200,000 face mesh and removed the few small polygons. Currently rebuilding the texture which takes about 35 minutes.
Update: Result is all black texture.
-
Hello David,
I think that such a long texture generation time is also caused by disabled interpolation (former Sharp mode). Using Interpolation should fix both long processing time and pixelated texture.
-
Using Interpolation solved the problem, though it didn't short the time to build the texture. The long time is probably due to the large number of photos.
Thanks for the help!
-
Is it possible to generate orthophotos from the dense point clouds only?
Martin
-
Hello Martin,
No. Orthophotos should be projected on the polygonal model, so Build Mesh is a required step for Orthophoto and DEM export.
-
HI all: about classify ground points.
I?m very interested in orthophoto generation from aerial view.
I want to achieve some kind of workflow to generate a point cloud and a mesh free fron buldings and trees to generate surfaces to generate contours of the real terrain.
I?m trying with the "tool classify ground points", but i?m stuck at this point.
My steps: align photos-build mesh Height field-create dense cloud.
There is something wrong here cause "the tools - Dense Cloud - Classify" is not activated......
Help, please (I?m working with a Trial License)
-
Hello bugonair,
Dense cloud classification is only possible for referenced chunks.
-
Hi Again:
You?re right. Now, with referenced chunk, What would be the process to delete items such as trees and buildings out of the model?
PD. Really this is the function in which I am interested
-
Hello bigonair,
Automatic dense point cloud classification finds points from the two classes: Ground and Low Points (that are under the estimated ground level). Manual classification allows to assign classes to the selected points.
It's recommended to use default values in Classify Ground Points dialog, but they can be slightly varied depending on the project.
Generally the method is the following: at first in every cell the lowest point is selected (Cell Size should be adjusted so that there are guarantied ground points in each cell). Then these points for each cell are triangulated in grid that should not be laying above the ground. Then new points are added to this grid - each of the points being added should match the following criteria:
the distance between the point and the "ground surface" should not be bigger than Max.Distance parameter; the angles between the "ground" and the vectors connecting the point and vertices of the corresponding cells are not greater then Max.Angle parameter.
Later in the Build Mesh dialog you can choose classes that should be used for mesh generation.
So once dense point cloud is reconstructed you should proceed to classification (Tools Menu -> Dense Cloud -> Classify Ground Points), if the default parameters do not provide desired result you should adjust the values (please note that prior to next run of classification you need to select all dense cloud points and assign to them "Never classified" class).
If the result is acceptable, you can build mesh based only on Ground Points class to get DTM.
-
Hello Ronald,
Could you please try build 1774, we have fixed some issues for orthophoto export there that may have appeared for complex objects (at the second issue should be fixed).
The strange colours still occur in build 1774 while exporting orthophoto using colour correction. It does not happen without colour correction. I checked my images and it seems to be some kind of footprint (better the shadow of a footprint) of some low exposed pictures. I am sorry I don`t have a log file.
Another bug I just experienced in build 1780 is the change image path command. It does not work properly. Changing image path will change the path if you check for the image path of the pictures but if you want to allign the chunk again it says images are missing and directs to the previous path. After saving the job and re-open photoscan the pathes are changed back to the previous pathes again. Forget about it....now it works...
-
Hello,
Can someone please tell me where I can find the fill holes option for opthophoto? Thank you!
-
Hello gheflorian,
Fill holes option is always enabled in the the version 1.0.0 for orthophoto export.
-
Then how come I have gaps in my ortho? I tired to export to 0.9.1 and with fill holes it worked just fine.
Look at the screenshot made with 1 and 0.9.1
-
Have the same problems with 1.0... So that for orthophoto generation I am using version 9.1. It takes longer but the result is better...
-
I also have the same problem with smaller gaps/holes.
Moving trucks/cars are also getting funny (see the attachment)!
-
With version 0.91 if i only need ortho i align photos on high and create point cloud typ mesh and i get good enogh result. Sometimes there are small stitching errors around roads but vegetation seems to look moore natural.
I tried this in 1.0, align on high and the generate mesh on high setting, this resulted in really badly stitched roads.
Is this quick way of doing orthophoto gone with 1.0 or was it just bad luck for me...
-
The roads became ok with dense medium cloud and high mesh. But what about theese holes, I have tried default and extrapolated (smooth) but they are still there.
-
What happens when I use the "rotate object" tool on a pointcloud after aligning it with GCPs? It looks like I move it relative to the un-anchored markers, and the XYZ axis stay still while the model rotates. I worry that I have changed my model orientation, but don't understand because the anchored GCPs stay in place.
Is this purely visual in a model where I am using a global coordinate system?
-
When interpolation is enabled in "build mesh" - what type of interpolation is used? It looks kind of like a spline interpolation?
-
@ andyroo: Poisson Reconstruction
http://research.microsoft.com/en-us/um/people/hoppe/poissonrecon.pdf
best regards
porly
-
What happens when I use the "rotate object" tool on a pointcloud after aligning it with GCPs? It looks like I move it relative to the un-anchored markers, and the XYZ axis stay still while the model rotates. I worry that I have changed my model orientation, but don't understand because the anchored GCPs stay in place.
Is this purely visual in a model where I am using a global coordinate system?
Hello Andy,
When you use Rotate Object instrument the model is being rotated relatively to the chunk coordinate system. So in case you apply it to the georeferenced chunk, the model will be incorrectly positioned in the real world space unless you apply referencing once again using Ground Control pane.
-
Thank you for the clarification, Alexey. It would be nice if the "undo" function allowed to undo model rotation - I was adjusting bounding box area and accidentally rotated the model - couldn't undo so I had to exit, reload. with a project getting up to 11GB it can be time consuming. :)
-
There are a whole bunch of things which should be undoable that are not...I feel your pain!
-
In the dense point clouds (or model/surface generation) is there any way to get rid of disconnected islands of point clouds, which then become shards in my model? PS 1.0 is generating a lot of erroneous surfaces above or below (mostly below) the actual surface, primarily in trees but also water, and it is maddening to edit all of these. Height-colored dense pointcloud in attachment. The big red areas aren't so bad, (they're areas of forested hillslope that PS "got") but the blue islands around the edge are forest that should be up high, but is instead at the base of the model.
I can provide the project file if needed, but it's 11GB unless I delete the dense pointcloud and model.
-
Then how come I have gaps in my ortho? I tired to export to 0.9.1 and with fill holes it worked just fine.
Look at the screenshot made with 1 and 0.9.1
Hello gheflorian,
Please check the latest update, we have fixed this issue with orthophoto export.
-
Hi Alexey,
Small request... Would it be possible to put the "Filter Photos by Marker" button back onto the Photos pane toolbar. I found it much easier and quicker there?
Thanks.
-
Then how come I have gaps in my ortho? I tired to export to 0.9.1 and with fill holes it worked just fine.
Look at the screenshot made with 1 and 0.9.1
Hello gheflorian,
Please check the latest update, we have fixed this issue with orthophoto export.
Thank you!
Indeed it's fixed now. I will report if I find more problems.
-
Hey All,
I've noticed that in 1.0, when I drag .tif files to the chunk it will import them in twice. So instead of pulling in say 10 pictures, I have 20 due to duplicates. However with .jpg and .png it works properly. Anyone else see this behavior?
Thanks,
Corey
-
Hello to all,
I am following the development of the pre-release version of 1.0.0 with all the innovations and new possibilitys very positively.
But unfortunetly I have still problems with the orthophoto generation. It is much faster than in 0.9.1 and color correction works fine.
But the comparison to 0.9.1 shows that the behaviour of the new kind of orthophoto generation works unfortunately worse.
For this process I always have to unistall version 1.0.0 and export the final photo with 0.9.1. Even that it takes longer but the result is always better.
For the comparison I used the model generated with version 1.0.0. So the conditions of each orthophoto are the same. There were no editing or decimating step for both photos.Please note that it wasnt my intention to produce a pretty orthophoto but to show the differences between these versions.
The image quality is not perfect but I choose this data to show how these versions handle situations like this. The result shows that the quality of version 0.9.1 is better in each case.
1.0. seems to have problems in the edge areas, what causes very noisy regions. 0.9.1 solves this issue "smoother".
I hope that there will be some improvements until the final release appears, since it is very laborious to switch between two versions during the process.
Best regards
Porly
-
some more
-
and more ;)
-
I am using the latest build 1789
-
Hello Porly,
Thank you for reporting. We'll try to improve orthophoto generation.
-
hello Alexey
recently upgraded my beta system to 1789 and found textures generated are only about 10% to 20% of a resolution comparing to 1760 build
mesh and texture built and exported on 1760
http://scr.hu/1rd/4oj2r (http://scr.hu/1rd/4oj2r)
http://scr.hu/1rd/tujm6 (http://scr.hu/1rd/tujm6)
mesh and texture built and exported on 1789
http://scr.hu/1rd/ijhyd (http://scr.hu/1rd/ijhyd)
http://scr.hu/1rd/qic71 (http://scr.hu/1rd/qic71)
no changes in options
please advise
augeaz
-
Hello augeaz,
Thank you for reporting. We've already been told about that and the fix should be included into the next update.
P.S. Just to make sure, the model has been rebuit in the latest pre-release?
-
Alexey
thank you for a prompt reply
steps taken in one of my tests before posting
1. build mesh, build texture in 1760 everything OK
2. open above project in 1789
3. remove texture, build texture (still OK)
4. remove model, remove texture, build mesh, build texture (low resolution)
building the model from scratch gives the same low resolution results as well though
cheers
augeaz
-
PhotoScan Version 1.0.0 build 1789 (64 bit) Windows 8.1 crashes and PhotoScan closes immediately when Align Chunks using Point Based method is used.
Align Chunks using Camera Based method works.
Version 9.1 build 1714 works with same files (saved as version 9.1 type) for both Align Chunks methods.
Crash reporter contents:
ARB_texture_non_power_of_two: Yes
ARB_vertex_buffer_object: Yes
OpenGLMaxTextureSize: 16384
OpenGLRenderer: AMD Radeon HD 7700 Series
OpenGLStereo: No
OpenGLVendor: ATI Technologies Inc.
OpenGLVersion: 4.2.12422 Compatibility Profile Context 13.152.0.0
ProductName: PhotoScan Standard Edition
System: Windows 64bit
SystemMemory: 16.0 GB
Throttleable: 1
URL:
Vendor: Agisoft
Version: 1.0.0.1789
XLog: OpenGL Vendor: ATI Technologies Inc.
OpenGL Renderer: AMD Radeon HD 7700 Series
OpenGL Version: 4.2.12422 Compatibility Profile Context 13.152.0.0
Maximum Texture Size: 16384
Quad Buffered Stereo: not enabled
ARB_vertex_buffer_object: supported
ARB_texture_non_power_of_two: supported
Loading project...
loaded project in 0.078 sec
Finished processing in 0.078 sec (exit code 1)
This report also contains technical information about the state of the application when it crashed.
-
Hello David,
Thank you for reporting. Point Based Chunk alignment in the build 1789 really leads to the crash, this issue will be also fixed in the next update.
-
Hey All,
I've noticed that in 1.0, when I drag .tif files to the chunk it will import them in twice. So instead of pulling in say 10 pictures, I have 20 due to duplicates. However with .jpg and .png it works properly. Anyone else see this behavior?
Thanks,
Corey
Hey Alexey,
Just another note on this, it only happens with drag n' drop, if the files are loaded directly through explorer they load just fine.
Thanks,
Corey
-
Hello
Often during the texturing (parameterizing texture atlas), the calcul stop at 70%...
(Only 40 Cameras, 1.9 Mo per picture) 64Go de RAM on my computer...
The project has 969566 faces and 859218 vertices. What append ?
When will be launch the final release of the 10 version ?
regards
-
Hello clardic,
What reconstruction parameters are you using for texture generation? If color correction is applied, the process will take significantly longer time.
-
Hello thanks for the fast reply
The texture generation was tested with standard parameters and 10000 x 1
-
Hello clardic,
And what about color correction in the advanced settings of the Build Texture dialog? Is it turned on by default? And what information is displayed in the Console pane?
-
Hello Corey,
Just another note on this, it only happens with drag n' drop, if the files are loaded directly through explorer they load just fine.
Thank you for the information. It may happen if the image files being imported via drag-and-drop have have several pages (for example, contain image thumbnails).
-
Hello Alexey
No correction color activated.
No real raison on other project, no problems.
When will be available the definitive version 10 ?
regards
-
Hello clardic,
Hopefully, version 1.0.0 will be released in a week or two.
-
Thanks for the ability to change the photo path within the interface. It's a life saver!
I'm excited about the new Estimate Image Quality feature as well. Where can I find information about which algorithm is used to determine the blur?
-
Thanks for the ability to change the photo path within the interface. It's a life saver!
I'm excited about the new Estimate Image Quality feature as well. Where can I find information about which algorithm is used to determine the blur?
What is the workflow to do that?
-
"Change photo path"
-
Hello eduardoestrada85,
Change Path option is available from the context menu by right-clicking on the photos' label in the Photos pane.
-
Hello
An other test with others pictures.
System seems to be blocked at 70% of calculation for texturing
No these problem before. Do I have to reinstall the system ?
A Screen copy of the calculation blocked
-
Hello clardic,
I recommend to open Console pane and watch the information in the pane during the processing - such information will be also informative for us to understand what is happening.
-
Please find enclosed the console information
-
Hello clardic,
Looks like the long processing time is the consequence of the huge texture charts number. Could you please provide us the original project file (with the original model!) for our internal tests? If the file is relatively small, then you can send it to support@agisoft.ru.
-
Hello eduardoestrada85,
Change Path option is available from the context menu by right-clicking on the photos' label in the Photos pane.
Tnx Alexey.... With this option is possible rebuild the texture process with other images?.
After align and build geometry with a first Images path?
Eduardo
-
Hello Eduardo,
It is possible if both of the following conditions are fullfilled:
- images have the same naming convention (otherwise, you'll have to change path for each image individually),
- corresponding images from the second set were taken from exactly the same camera positions.
-
tnx for the reply
The second set must be the same size en pixels?
The second set must be the same number of images, like the first?
Eduardo
-
Hello
Come back on the problems I haved.
It seems that the soft have had a indertermined bug.
I have desinstalled Photoscan. Reinstall it Same problem.
I have desinstalled Photoscan an cleanup the register basis by regedit of all topics about photoscan and agisoft. Restart the system, reinstall the soft, enter the licence Key. And now, all is ok !!!
Where is coming from ???
-
Hello clardic,
Do you mean you've only started texture generation process on the same project file you have?
-
No, I have restarted all the process from the begining.
But, you have to know that, before registred cleaning and reinstallation, I had tested to restart the generation from the begining of all the process without success.
It seems that a problem was arrived after some modifications in the parameters ( I tested the differents rsults gived by the soft when we changed parameters).
-
Hello clardic,
From the project file we can see that you've used disable interpolation mode and it results in a huge number of texture charts (over 24000). Enabling Interpolation or using Extrapolated method will definitely reduce the number of texture charts.
-
Thanks for your real support, I appreciate it, it's not so often.
I'll follow your recommendations.
I really appreciate your soft.
I'm waiting with big interest, the news tuto for the version 10.
Sometime when we have white part in pictures, we obtain some holes in the mersh? What do you suggest to avoid it ?
-
I had a strange outcome with PS 1.0 build 1789 when trying a new model-building strategy (build model from whole flight instead of chunks, with no interpolation). The project is 809 cameras (777 aligned) in 4 overlapping flight lines along a 30 km river corridor.
Normally I would align all images and add GCPs, then optimize alignment before splitting into chunks and building dense cloud and surface model with ~50 000 000 faces, with interpolation enabled. This time I decided to build a single surface with no interpolation (Was hoping to avoid hole-filling in wetted channel). I ended up with a very broken-up model with poor detail resolution.
First I aligned all images and added GCPs, then optimized alignment. Then I built dense cloud and mesh in batch process. My dense cloud is 622 613 930 points, but the model only has 12 096 826 faces. There is virtually no detail in the areas where normally I can see quite a lot (terraces, logs, etc). I attached an image showing the sparse point cloud with the model generated from the dense cloud in the same area (sparse cloud is only 832 012 points, but still better resolution than the model.
Dense point cloud reconstruction was 3 hours 33 minutes, and model generation was only 34 minutes 43 seconds. What I don't understand is that there were clearly a lot of points in valley bottom areas (even in sparse cloud generation), but these points look like they were "skipped" for surface reconstruction. I don't know if that's because of the "aggressive" setting or the "no interpolation" setting but it seems weird. There are holes in the surface model where the river is, like I expected, but in the valley bottom, where the detail should be good, it is very poor - only a few polygons instead of thousands.
It would be nice if instead of interpolation/extrapolation/disabled, I could specify maximum area (or distance) to interpolate, and some way to exclude the small islands outside of the main model area.
Settings were:
Build Dense Cloud
Quality: Ultra-high
Depth Filtering: Aggressive (read in a post that is preferred "default" setting now?)
Reuse depth maps: No
Build Mesh
Surface Type: Height field
Source data: Dense cloud
Interpolation: disabled
Face Count: Custom
Custom Face Count: 50000000
-
Hello Andy,
And have you tried to reconstruct mesh with enabled interpolation setting based on the same dense cloud?
-
Hello Alexey
I can see PS 1.0.0 came out of beta but I still have to keep it on my testing machine as the problem with textures persist, unless I am doing something wrong.
http://scr.hu/1rd/kn9c3 (http://scr.hu/1rd/kn9c3)
http://scr.hu/1rd/ufsut (http://scr.hu/1rd/ufsut)
http://scr.hu/1rd/ou1dx (http://scr.hu/1rd/ou1dx)
http://scr.hu/1rd/76voe (http://scr.hu/1rd/76voe)
Everything was done from scratch.
1. add photos
2. build dense cloud
3. build mesh
4. build texture
5. export texture
A very sparse texture
http://scr.hu/1rd/4hhua (http://scr.hu/1rd/4hhua)
Please advise.
Regards
Marcin
-
Hi Marcin,
I got the same result of yours when I use V.1780 few weeks ago.
After that I downgrade the version to the one before 1780.
I am not sure does the 1789 and 1795 already fix it or not.
Will try later.
BTW it is not a always problem. Some results are good.
Blmwnew
-
just tested the 1795 build, and the texture is still VERY sparse...
-
Hello,
In the latest builds mesh generation method in Arbitrary mode has been revised to produce sharper details of the geometry. The scattered texture atlas with huge amount of texture charts is caused by these changes. We are currently checking the possibility to improve both stages (reduce artifacts on geometry mentioned by AlexanderT and improve texture atlas parametrization) and if everything goes fine improvements will be implemented in the 1.0.0 update.
-
Hello!
Im not sure if this is a bug or if it was reported. The issue I am having with, is exporting orthophoto image - top down view of an archeological excavation. Exported image has an overall heavy greenish tint, losing the correct colors that can be seen on the models texture. The function autocorrect color (on or off) produces the same results. Also it matters not whether blending mode is mosaic, average or max min. intesity.
edit: I am using the latest build of the pre-release
-
I'm looking for more information about the algorithm used for the "Image Quality" feature. I would like to know how it determines blur. Thanks!
-
Hello Kerri,
PhotoScan estimates the sharpness of the borders between separate areas on the photo and according to this information provides Image Quality value.
-
Hello Zissou,
Can you please post several screenshots describing the problem: the resulting orthophoto, the overview of the model with and without camera positions?
-
Hi Alexey,
The latest 1795 is same to 1780 that the texture is still VERY sparse and some time the texture is whole black.
Would you please try to correct this ASAP?
Currenly we can only use older build for our work.
Best wishes
Blmvnew
-
I'm sure Agisoft folks are doing their best to resolve the issue, and in the mean time, we can roll back to the latest build before the new algorithms were introduced. If I am not mistaken, it is build 1774, correct?.
I have a question though: I already have a lot of projects done on latest builds, with mesh artifacts similar to those everyone else is getting. After rolling back to older build, do I just have to rebuild mesh to get good old clean geometry results? Or do I have to rebuild dense point cloud on that build as well?
-Andrew
-
Hi Andrew,
By my experience the dense point cloud that build by the latest version can be used in old build and the result is ok.
Best wishes
Blmvnew