Agisoft Metashape
Agisoft Metashape => General => Topic started by: Dmitry Semyonov on July 09, 2012, 10:23:05 PM
-
Pre-release version of Agisoft PhotoScan 0.9.0 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 0.9.0 version is released.
Note
Project files created with 0.9.0 version can't be opened with earlier versions.
EDIT:
PhotoScan 0.9.0 build 1584 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)
-
Hello Agisoft Team,
After making my first tests. I have mixed feelings. On one side is very important changes, but on the other side quality of the model is lost, for example, version 0.8.5 performed in an extraction 11,338,164 vertices with the same parameters, version 0.9 performed in an extraction 1,731,924 vertices, And this is clearly visible in the model.
The loss of quality of the model is visible in the new version. I hope you can correct.
With the explanation of Alexey, the solution is clear
Best Regards,
Diego
(http://)
-
How come these pre-releases can not be installed by the side of the stable versions on Windows... It is very easy on the Mac to have many versions of Photoscan at the same time.
As you wouldn't want to change to a pre-release only, but still would like to access it and try out.
-
try create another USER accound under Win and install there.....
-
Hello Diego,
In 0.9.0 zero face count for Height Field objects now has different meaning. Now if "0" is used in Build Geometry dialog it refers to the default (recommended) value and not to the maximum number of faces.
So if you require more faces you should specify face count to desired or exceeding value.
-
Hello Alexey,
Thanks for your explanation, you're very kind.
As always, excellent support.
Best Regards,
Diego
-
Hello Dimitry (or Alexey)
when looking at the change log.. I don't see the 0.9 version changements....
Thanks
f/
-
Agisoft PhotoScan Change Log
Version 0.9.0 build 1530 (10 July 2012, preview release)
Standard and Professional editions
• Added support for OpenEXR image format.
• Added support for HDR texture generation.
• Added Import Cameras command in Batch Process dialog.
• Added Align Chunks command in Batch Process dialog.
• Added Add Folder command for batch image import.
• Added support for automatic mask generation based on background photos.
• Added ability to store and reuse calculated depth maps.
• Added Mask and Depth Maps viewing modes to Photos pane.
• Added Reconstruction uncertainty criterion for point cloud in the Gradual Selection dialog.
• Added Fill Holes option in the Build Texture dialog.
• Improved texture quality in Average blending mode.
• Improved performance in Height-field Smooth reconstruction mode.
• Improved performance in Arbitrary Smooth reconstruction mode.
Professional edition
• Added Timeline pane and support for 4D dynamic model reconstruction.
• Added support for multispectral orthophoto generation.
• Improved orthophoto blending quality in Mosaic mode.
• Added support for coordinate system based on Hotine Oblique Mercator projection.
• Added Track Markers command for multiframe chunks.
• Added support for processing report generation.
-
"Added Timeline pane and support for 4D dynamic model reconstruction."
I downloaded the new version but don?t know what this does.
And what is the 4D dynamic model reconstruction?
-
oups Thanks Diegotorres... I had forgot to refresh the page... ;D
-
"Added Timeline pane and support for 4D dynamic model reconstruction."
I downloaded the new version but don?t know what this does.
And what is the 4D dynamic model reconstruction?
http://www.ir-ltd.net/update-and-4dtimeline-plus-difference-masking-in-agisoft
-
Well, I tried with a set of my photos but it didn?t work. I don?t know how to make it work.
-
Hello Fabricio,
To create multi-frame chunk you need to organize images by folders and then use Add folder command from the Workflow menu.
For example, each folder corresponds to camera and each image in these folders - to the frame. So you'll have several "Camera##" folders containing "Camera##-frame####.jpg" images (naming is an example, it could be different).
-
try create another USER accound under Win and install there.....
That will not work, and it wouldnt be very useful ether..
-
Hello Diego,
In 0.9.0 zero face count for Height Field objects now has different meaning. Now if "0" is used in Build Geometry dialog it refers to the default (recommended) value and not to the maximum number of faces.
So if you require more faces you should specify face count to desired or exceeding value.
It would be very handy to still have the option of specifying "do not decimate" with some value or checkbox. Like it used to work with using 0.
How does the recommended value work anyways, does it take into account the Maximum RAM in the machine to make sure it will not cache to disk, or what other vars. are there?
-
After conducting various tests. I prefer the concept is handled as in version 0.9.0, it is much better. Great work team agisoft, Thanks.
Best Regards,
Diego
-
Hello olihar,
Now the model will not be decimated if you specify the intentionally bigger values. Currently PhotoScan can handle approximately up to the two billion faces.
If the face count in Height Field is set to zero the triangulation process continues until the difference between values in DEM grid and triangulated values are less then half of a grid step.
-
After a few tests, I find this version very stable, more effective in resources use and the DTM's are looking great. It seems that the resulting orthophotos are better than in version 0.8.5. Nice work!!
Cheers
-
I have a couple of questions about how images formats are handled in 0.9.0
I have some 4 band digital aerial images in tif that now open in 0.9.0 without artifacts, but don’t show as color. Is there a way to change band combinations?
I also have some scanned color aerial tif images that now open in 0.9.0 as black and white also.
Curious.
Tom
-
Hello Tom,
When multi-layer TIFFs are loaded into PhotoScan only one layer will be shown and all the calculations (Align Photos, Build Geometry) will be performed by the only layer (Master channel). Master channel could be selected from the Chunk context menu in the Workspace pane using "Set master channel" command.
But when you export Orthophoto to TIFF from such project the resulting image will also have the same number of layers as original images (+ alpha channel).
-
Hello Alexey,
So in a normal 3 band RGB image that is displayed in color, only shows one channel under "Set Master Channel"
However, I have some images that only have 3 bands, 8 bits per channel, really a RGB image, but the file info shows a property -
Photometric interpretation YCbCr
and therefore has 3 channels, under "Set Master Channel" and will only display and use the master channel.
And the sparse point cloud from align photos as well as the resulting geometry only shows the master channel.
I think there needs to be a way to choose the display band combination and not be limited to a single band.
Tom
-
OpenCL does not seem to be working at all... Previous versions I could see GPU activity in my ATI driver window. Now it stays at 0%. I have the preferences set correctly (2 cpu cores disabled). I am using a 4.7GHz 8 core AMD FX-8150 and 2 AMD Radeon 6970 2GB cards with 16GB system memory. Let me know if this is a known issue and is being addressed. Thanks!
-
Hello tweezlednutball,
Could you please specify if this problem appears in standard workflow or during Batch processing? Also please tell us if the GPU device is successfully recognized in PhotoScan Preferences window and checked on.
-
Standard workflow. The 2 devices are checked and 2 cpu cores disabled. :'(
Photoscan seems to be disabling the cores properly as its using 75% of my processor which would be 6/8.
I am trying batch now to see if that makes a difference.
-
Hello tweezlednutball,
And could you please provide the log file for the simple project showing build geometry stage.
-
Tried batch mode. Not working either.
-
I guess it was a false alarm. I restarted my computer and now I am getting activity on my graphics card. :D
Maybe you should incorporate a restart after the installer completes? Thanks so much for the help.
-
OK, thanks for informing.
We'll check if we are able to reproduce this problem and if it could be solved by restarting.
-
How do you export the undistorted photos in 0.9? No matter what combination of path I type in it responds with, "Please specify the destination folder"
What do I need to type in to get them to properly export again?
-
Hello Maharg,
Thank you for reporting.
The problem is fixed. Please refer to build 1534.
-
Brilliant! You guys are the best!
-
Is there any noticable speed differences etc with standard edition and professional edition ?
Cheers
Glen
-
OpenCL
I am presuming the openCL drivers come with nvidia drivers as standard and AGI automatically picks up your device in the preferences ?
Do I have to install seperate openCL drivers or anything like that?
I have a single GTX680 OC 2GB card.
So turning openCL on and setting 7/8 cores would be the correct setup ?
Im still not sure if there is a noticable improvement with a scene yet. I am still running more tests.
-
Glen can send us the benchmark of the 680 card ? run a scene with 3 images, with HIGH setting and ARBITARY and SMOOTH, the see in the console for the SAMPLES/SEC of the 680 card....
-
Hey WG
I am running
i7-3820 @3.6GHz ..overclocked to 4.0GHz
32GB RAM
win7 64 version of 0.9.0 pre release
3 images and set to High Arbitrary Smooth (100,000 polys)
Device 1 performance: 118.964 million samples/sec (CPU)
Device 2 performance: 199.001 million samples/sec (GeForce GTX 680)
Total performance: 317.964 million samples/sec
-
This is pretty interesting... I think I will stick to the GTX580, the 680 cards seems to be driver crippled as people thought.
3 Images same settings as you used.
Device 1 performance: 166.105 million samples/sec (CPU)
Device 2 performance: 293.831 million samples/sec (GeForce GTX 580)
Total performance: 459.936 million samples/sec
I have
i7-3930K over-clocked to 4.5GHz
64GM RAM
MSI N580GTX Twin Frozr II/OC
win7 64bit
Such a shame of Nvidia to do this.
Then moving up to Ultra on these 3 particular image I got about: 333 million samples/sec (GeForce GTX 580)
Hey WG
I am running
i7-3820 @3.6GHz ..overclocked to 4.0GHz
32GB RAM
win7 64 version of 0.9.0 pre release
3 images and set to High Arbitrary Smooth (100,000 polys)
Device 1 performance: 118.964 million samples/sec (CPU)
Device 2 performance: 199.001 million samples/sec (GeForce GTX 680)
Total performance: 317.964 million samples/sec
-
Hello Oli, it seem realy asim mentioned earlier, have few benches for 580 and even the 590 this is first of the 680. can anyone post the 680 test data , im preparing a google docs table with data for all......
-
nutz !
I just sold my other cards too. I should have kept them. Perhaps NVidia will post a driver update soon for 680 ?
-
Heres a quick bench with the openCL turned off.
Device 1 performance: 168.758 million samples/sec (CPU)
Total performance: 168.758 million samples/sec
-
Hello glennn,
Is there any noticable speed differences etc with standard edition and professional edition ?
No there is no difference in processing speed between Standard and Professional editions (unless Ground Control pre-selection is used at Align Photos stage).
I am presuming the openCL drivers come with nvidia drivers as standard and AGI automatically picks up your device in the preferences ?
Do I have to install seperate openCL drivers or anything like that?
Installing up-to-date drivers for your graphic card is usually enough for PhotoScan to recognize OpenCL device.
-
Hello all,
Checking GPU performance in PhotoScan in the way mentioned above is not really correct, since it depend on different factors. We have tested GTX 680 by "pure" PhotoScan performance tests and the results are nearly equal to the result of GTX 580.
To check the performance experimentally you have to process the same dataset (with the same quality and bounding box selected) and also to disable all CPU cores. Thus the comparison will be more adequate.
-
I have been attempting Ultra High mesh generations recently with a new computer and have been running into the memory error. It seems that the reconstructing depth step is using over 4 gb of memory but when it goes on to generating meshes it ramps down to 4 gb max, uses only 4 cpus on the calculating, no gpu, and then fails with a memory error.
Though this is just speculative...does generating meshes just x86 itself all of a sudden?
I have 32 gb of ram with a Radeon 7970. The open cl is working fantastically on reconstructing depth.
-
Hello Maharg,
GPU devices are not used during generating mesh step of the Build Geometry stage.
How many images are you processing and of what resolution? And what geometry/object type are specified in Build Geometry dialog?
And please make sure you have 64bit version of PhotoScan installed.
-
Hey Alexey!
I disabled most of my images...only 11 are active for this generation. I have only used the 64 bit version, in fact reconstructing depth goes well above 8gb, but then generating mesh goes up to 4.3 ish then crashes out. I also double checked my ram with prime and everything checked out.
Thank you,
G
-
Hello Maharg,
Does PhotoScan crashes or just fails to build geometry with error message? Maybe the bounding box could be reduced?
-
I think it was due to me disabling the pictures instead of removing them. It's maxing out at 32 gb now. I'll let you know when I wake up if I have success.
Thank you.
-
Dem export gives a pure black and white (just) image. Am I missing something? On the contrary the report contains a colorfull DEM. By the way, I'd suggest to add a map for errors in control points.
some ideas:
Plot ellipses onto GCPts with axes proportional to x and y errors and colored by elevation error
Create a vector field plot for xy errors and overlay a "temperature" map showing z error (signed).
Regards.
JMR
PD. Great moves in 0.9.0!!!
-
Hi LFA
Depending on the application that you are using, you must find how to strech the DEM histogram for the Maximum and Minimum z values. Usually you can do so in the visualization options for the DEM file. Right click in ArcGIS and access the properties for the layer or, if you are using QuantumGIS for instance, you have a "Strech Histogram" button in the button bars.
Cheers
-
Hi Wishgranter,
I am in the process of finalizing another dual CPU number cruncher. Any chance of a look at the benchmarks for all the graphics cards in that Google doc. I have had a play with the 470, 570, 590 and a brief encounter with the 680. Thinking of getting a few 670's but it would be nice to pick an informed sweetspot.
Thanks
-
Welcome
I tested the old versions of the demo and the program has everything.
Yesterday I bought a new version 0.9.0 of the program.
I program it entirely is lost, he can not even submit a test project
My computer dual Xeon 24 core. GTX 580
Calibration sometimes worse results , and sometimes it's better.
The program is unpredictable ...
They are also pluses. Mesh generation is accelerated.
(http://desmond.imageshack.us/Himg191/scaled.php?server=191&filename=erroros.jpg&res=landing)
-
comparison 0.85.0 0.9.0
exactly the same settings
the new version works, very strange
as for me, the quality deteriorated
is a big problem with the function align
(http://desmond.imageshack.us/Himg28/scaled.php?server=28&filename=comparea.jpg&res=landing)
I've done a dozen tests on different images.
Always the same settings.
always well-calibrated camera
And the new program works very bad
-
well.. last time, as I was trying to align an old project, Pscan failed... I was very surprised... so I tried to re-align it a second time and I switched the lower panel from Photos to info log... with the same settings.. and it worked..
the only thing I did between the two was that I switched the photo panel to the info log.. and it worked.. but it can't be that right..? that be weird.. although who knows...
my0.1
f/
-
Hello kris3d,
Thank you for reporting.
Do you experience problems with alignment on different projects?
-
so are the problems in the various stages.
Connecting the camera from different angulation is very problematic.
The new program completely lost where the old program works beautifully.
When scanning the facade, the new version works fine but clearly you can see lower quality.
Scanning the interior finishes, a total failure
The align the new version brings more points.
But it does not give better quality.
Generating mesh run faster. Sometimes computation for multiple cores.
But there are situations where in the process created a sudden very large memory requirements.
Please check your example monument.zip
I'll do some other tests for other lenses
sending a test scene,
photographs + camera calibration
https://www.wetransfer.com/dl/b46Zpz4P/ea7a5c7dbb1a1001a47a786fdca22e31fb16e564a72daf1e737d406a3981b568ecc4d7b74d145d4 (https://www.wetransfer.com/dl/b46Zpz4P/ea7a5c7dbb1a1001a47a786fdca22e31fb16e564a72daf1e737d406a3981b568ecc4d7b74d145d4)
(http://desmond.imageshack.us/Himg703/scaled.php?server=703&filename=testybo.jpg&res=landing)
-
Thanks for providing the data, we will check it shortly.
By the first look using "Fix calibration" flag works fine in both versions, while autocalibration provides misaligned camera positions in both versions too. Could you please check you are using "Fix calibration" feature in 0.9.0?
-
I never used the option fix.
I did a test now enabled fix.
Scene looks much worse
(http://desmond.imageshack.us/Himg24/scaled.php?server=24&filename=fixsi.jpg&res=landing)
-
Hello kris3d,
We've looked at the project.
Our workflow was the following:
- adding photos, importing camera calibration data,
- aligning photos at High accuracy (Fix calibration disabled, since you say you do not use it),
- after alignment img_016 is wrongly positioned, so we used Reset photo alignment feature on it (using context menu in the workspace pane), and applied align selected again for this image - photos are positioned goo now,
- optimized photo alignment (this step could be skipped in the Standard edition),
- build geometry.
the project file (with the steps saved as chunks) is available by this link (ftp://agisoft:agisoft@ftp.agisoft.ru/home/downloads/compare-090.psz).
-
Alexey Thanks for the information. But it does not solve the problem.
Old versions 0.8.3, did a good fit cameras automatically.
So they were more intelligent :)?
For now what I have manually reset the camera?
How do I know which is the wrong fit?
This is a simple project for five cameras.
(http://desmond.imageshack.us/Himg191/scaled.php?server=191&filename=erroros.jpg&res=landing)
In a very complex projects, from 100 upwards camer.
I can not imagine manually adjust the camer.
Look at my first post, in the same project is a 34 camera
And everything fell apart, a newer version could not.
I tried to re-convert the old projects
and the effects are terrible
Soon sending more complex scene and try to do it.
good luck
Link to photos
https://www.wetransfer.com/dl/21RXevJ8/90b3ba607bb7dffa11c2a6916b6fdaeb4f3ccd7553d1821f91647c59d02d1358d7e466384d281cf (https://www.wetransfer.com/dl/21RXevJ8/90b3ba607bb7dffa11c2a6916b6fdaeb4f3ccd7553d1821f91647c59d02d1358d7e466384d281cf)
link to the scene in the picture
https://www.wetransfer.com/dl/dK1vv29T/7d6c917f59489478c634e8a71777c8563aeae565b9abafec6f6e441a35fc4fdc8bf120a94d2cc9c (https://www.wetransfer.com/dl/dK1vv29T/7d6c917f59489478c634e8a71777c8563aeae565b9abafec6f6e441a35fc4fdc8bf120a94d2cc9c)
-
Alexey Thanks for the information. But it does not solve the problem.
Old versions 0.8.3, did a good fit cameras automatically.
So they were more intelligent :)?
:D more intelligent ;D
The problem is you, my friend. :o
The calibration parameters of the camera is bad. <cx> and <cy>, are reversed. Image rotation to 90 degrees.
(http://i46.tinypic.com/24gv9d3.jpg)
I do the whole process with the corrected camera, and runs smoothly
(http://i49.tinypic.com/21llbwl.png)
Here you can download the version 0.9 project http://wtrns.fr/0mMYUO1mTvJjq0 (http://wtrns.fr/0mMYUO1mTvJjq0)
If the parameters of the camera is good. Alignment works well. Version 0.9
(http://i47.tinypic.com/15pq1xd.png)
Align OK Version 0.9 = http://wtrns.fr/GkaPMDVf3-nb_rM
Best Regards,
Diego
-
can not comprehend that my friend diegotorres
But if it works then you're great :)
Tomorrow I'll try to understand it.
How not to understand.'ll Ask you to :)
thanks
-
Is there a changelog for the 1556 build?
-
Agisoft PhotoScan Change Log
Version 0.9.0 build 1556 (05 September 2012, preview release)
Standard and Professional editions
• Bug fixes.
• Gradual Selection.
I think so ... ;D
-
is version 0.9.0 build 1556 standard edition available to download ?
I'm still on 0.8.5.1423 and the updater says no updater required
confused or am i missing something ?
cheers,
mala
-
Hello mala,
PhotoScan 0.9.0 is currently in pre-release state, that's why you do not receive messages about new version.
But you can try it downloading the latest build (1561 already) from the first post of this topic.
-
Ah silly me! didn't think of looking at the start of thread.
thanks Alexey and keep up the good work!
-
Hi WishGranter
I have been benchmarking the new NVidia Drivers 306.23
previously on the 304 drivers i was getting this result.
I am running
i7-3820 @3.6GHz ..overclocked to 4.0GHz
32GB RAM
win7 64 version of 0.9.0 pre release
3 images and set to High Arbitrary Smooth (100,000 polys)
------------------------------------------------------------------------------------------------
304.79 drivers
Device 1 performance: 118.964 million samples/sec (CPU)
Device 2 performance: 199.001 million samples/sec (GeForce GTX 680)
Total performance: 317.964 million samples/sec
------------------------------------------------------------------------------------------------
306.23 drivers
Device 1 performance: 159.302 million samples/sec (CPU)
Device 2 performance: 265.46 million samples/sec (GeForce GTX 680)
Total performance: 424.762 million samples/sec
-
how do I use this new feature?
Added support for automatic mask generation based on background photos.
Many thanks
Stefano
-
Ok now I understand, excuse me
Stefano
-
Hi WishGranter
I have been benchmarking the new NVidia Drivers 306.23
previously on the 304 drivers i was getting this result.
I am running
i7-3820 @3.6GHz ..overclocked to 4.0GHz
32GB RAM
win7 64 version of 0.9.0 pre release
3 images and set to High Arbitrary Smooth (100,000 polys)
------------------------------------------------------------------------------------------------
304.79 drivers
Device 1 performance: 118.964 million samples/sec (CPU)
Device 2 performance: 199.001 million samples/sec (GeForce GTX 680)
Total performance: 317.964 million samples/sec
------------------------------------------------------------------------------------------------
306.23 drivers
Device 1 performance: 159.302 million samples/sec (CPU)
Device 2 performance: 265.46 million samples/sec (GeForce GTX 680)
Total performance: 424.762 million samples/sec
These are very inaccurate test results. just looking at the increase in CPU as well... One would have to run these tests at least 10 times to get some results that can be used to determine improvement.
I do wonder if these new drivers work with 580 card as 2 last updates completely crippled these 5xx cards...
-
There seem to be a problem with camera export and import in this new version, I get all kind of strange results with it. Both the xml and the bundle.out file.
I think the main problem is in the export.
-
Hello olihar,
Could you please describe the problem. When you import cameras back does anything change in calibration dialog or in model view (camera positions)?
-
At least in my last test, I exported cameras, then tried reimporting and do matching again like it asks and it made a complete mess... I need to try it out again...
There might be a reason for it, if you have some photographs run the bundle adjustment, then you delete some photographs and export again, the order of the images might get strange... I wonder maybe worth checking out... just some quick thoughts as I need to try it out again.
-
I actually tried again, 300 images in ran all the things needed, exported cameras reopened images and imported the file and things where completely wacked out.
-
I am getting the same photo order reshuffle issues. Is there a way to reorder the photos or do you hav e to start the project from square one again. :(
-
Hello Matt,
Could you please describe how we could reproduce this problem? What are the steps and parameters used (for example, if chunk is referenced)?
-
Hi Alexey, I have been trying to avoid doing it actually ;D. It was done merging two referenced chunks over the same area taken a year apart. It seems to have happened when I merged the chunks based on the two lots of ground control then did a fresh alignment. One of the strips had lots of overlap >3.5 and the other heaps of shadow (bad combo). I get a bit if the crazy points off at weird angles but just clip them out and i can still work with the outputs.
-
HI Alexey,
After a bit more investigating i think it results from camera calibration stuffup. But once the order gets shuffled you cant get it back. See Image
-
In the "Agisoft PhotoScan Change Log" I found today:
Version 0.9.0 build 1584 (7 October 2012)
Standard and Professional editions
• Added support for model export in FBX format.
• Added Polygon size criterion for mesh in the Gradual Selection dialog.
• Added free-form selection tool.
• Added Depth filtering option in the Preferences dialog.
• Added Calibration mode option in the Camera Calibration dialog.
• Added From Model option in the Import Masks dialog.
• Added anaglyph rendering mode for consumer grade video cards.
Professional edition
• Added support for automatic coded target detection (12 bit and 16 bit).
• Added support for marker pinning.
• Added support for C-Astral Bramor telemetry log import.
• Added Split... command for multiframe chunks.
• Added automatic zooming on selected marker in the Photo view.
Really very nice new features! But from where we could download it? ;) On the first page of this topic there is still only version 1573. Any informations on the new features are also welcome!
Best regards
Karsten
-
Karsten, go ahead and DL it, it is the build 1584... 8) even though it says 1573
f/
-
• Added Polygon size criterion for mesh in the Gradual Selection dialog.
• Added free-form selection tool.
• Added Depth filtering option in the Preferences dialog.
haha! These changes sound absolutely great! Thank you so much!
-
New released version 0.9 since yesterday at normal download.
Thank You Agisoft!
Karsten