Agisoft PhotoScan 1.1.0 pre-release

[StevenF]

Hello Steven,

Thanks for reporting, however, we haven't been able to reproduce the problem with crashes.

But as for the creating duplicated copies of chunks and adding them to the existing document, we will try to implement this functionality in Python API.

Hi Alexey,
It could be my install. I'm working with the 64-bit Linux version which fails on the chunk.copy() command. I just downloaded the Windows 64-bit version and it worked fine. I load in a project and execute the following at the console:
import PhotoScan
doc = PhotoScan.app.document
chunk = doc.chunk
chunk2 = chunk.copy()

[Alexey Pasumansky]

Hello James,

I've just checked color correction option and it took much longer time in the latest pre-release than export without any correction. Do you have any log information saved related to the both processes?
Even in the version 1.1.0 there were some optimizations for the color correction (speed up gained on the AVX supported systems), the procedure still requires quite a lot time.

[andyroo]

I have had good success running a few jobs generating orthos and DEMs in batch mode and setting coordinate systems with build 1985. This is very nice for the color correction step, since it takes several hours on my machine with 1000 - 2000 images.

One thing still lacking is ability to control cell center/corner coordinates. This is critical to generate concurrent rasters (exactly overlapping cells), which is important for accurate surface comparison.

The simplest way that I can think of to implement that would be to round the DEM extent/corners to the resolution of the DEM (so that if for instance I select a 1 m cell size, the bounds would be rounded to the nearest integer, if I select a 0.5 meter cell size, the bounds would be rounded to the nearest 0.5 m, if I set a 0.1 degree cell size my bounds are set to the nearest 0.1 degree... etc.

The default resolution (0 in current beta) could be left as a way to not implement any rounding and use the maximum resolution from the data.

The other way I can think of would be to add an option (like a checkbox + value field) to specify the rounding interval for cell corners. If the box is unchecked, then default bounds would be used. Either of these options don't seem like they would limit current functionality at all.

[andyroo]

One other thought on batch processing, orthos, and DEMs - when the user selects "save project after each step" currently Photoscan saves after any batch step is complete.

This  doesn't seem necessary for batch "export" actions since they don't modify the project. In my case it can take around 30 minutes to save a project, so it would be more efficient to not save after export ortho and DEM.

Andy

[augeaz]

Hi Alexey

What are the chances of having an option to export cameras in batch same as import option please.

Regards
Marcin

[stpu]

Hi,

will it be possible in the final version to export undistorted images with more than 3 bands?

Thank you.

Stefano

[ppkong]

There are some problems in version 1995, the texture looks like it has some errors.
Another question about AgiSoft Viewer,When I save a playlist and reload playlist(Open Playlist),I can't see anything untill I press F5.

[Alexey Pasumansky]

Hello Andy,

We'll try to add this feature in the next update. As for the autosave operations, you can use Save task in the Batch Process dialog.


Hello Martin,

We'll check if it could be easily implemented for the next update.


Hello Stefano,

Next update will include the fix for the correct undistort of the multi-band files.

[Alexey Pasumansky]

Hello ppkong,

Maybe you can provide the project file for the further problem studying?

[kenwoods1]

We have processed a set of aerial photos using Photoscan Pro Linux version and the same photo set with Photoscan Pro Windows version with completely different results. The main issue is that a large sections of the survey area was not constructed during dense point cloud generation in the Linux version, but was constructed completely and accurately in the Windows version (see attached screenshots). We went through many interations of processing, making subtle adjustments to the workflow, but in the end achieve the same unsatisfactory result in on the Linux system. The Linux DPC results makes it impossible to generate accurate DSM and orthophotos. 

The standard photoscan workflow was used in both projects. Camera position coordinates are in WGS84 (EPSG 4326).  Latitude = column 4, Longitude = column 5, Elevation = column 6. We used a 20 cm camera accuracy parameter and optimized with the default settings after running a high-quality alignment using our camera reference coordinates. After unchecking a few photos with high positional errors, we re-optimized and got total errors under 15 cm.


I can supply the images and more details----There's 1500, 23MB jpgs.

[andyroo]

I discovered an interesting issue with alignment that affects builds 1985 and 1995. I aligned photos, added control, then optimized alignment and saved/exited.

Later I added more GCPs at one of the ends of the flight and re-optimized, but I wasn't getting any change in calculated error of GCPs.

I saved the project as a version 1.0 file and ran the optimize step in 1.0 and the images were realigned to minimize the error (from 19 m and 13 m on the two additional GCPs down to 0.2cm or less), but when I ran the optimize step on the 1.0 file with 1.1 build 1995, there was no change.

I don't know why it's not further optimizing. Can share entire project or just project files.

[James]

I discovered an interesting issue with alignment that affects builds 1985 and 1995. I aligned photos, added control, then optimized alignment and saved/exited.

Later I added more GCPs at one of the ends of the flight and re-optimized, but I wasn't getting any change in calculated error of GCPs.

I saved the project as a version 1.0 file and ran the optimize step in 1.0 and the images were realigned to minimize the error (from 19 m and 13 m on the two additional GCPs down to 0.2cm or less), but when I ran the optimize step on the 1.0 file with 1.1 build 1995, there was no change.

I don't know why it's not further optimizing. Can share entire project or just project files.

I had the same problem, i learned that it was because there were some cameras in the reference pane which had 0 projections, as a result of either manually removing points from the sparse cloud or using the gradual selection tools.

you can sort the cameras by number of projections (it's the rightmost column and often not visible without scrolling horizontally), then select those cameras with 0 or very few projections, and either remove them or align them again (using align selected cameras).

[chrismaclean]

Hi Guys,

Testing the 1995 release and  OpenCL hard crashes my solve box when using it.  OpenCL works fine with 1.0.4.  Leaving OpenCL off in 1995 everything runs fine.

*Running Centos 6 and two Titans.  Is there a specific OpenCL version I should be testing with for Linux?

Thanks.

[andyroo]

I had the same problem, i learned that it was because there were some cameras in the reference pane which had 0 projections, as a result of either manually removing points from the sparse cloud or using the gradual selection tools.

I have one (disabled) camera with one projection, and a fair number of un-aligned cameras that are blank for projection, but nothing showing 0 projections. Since this behavior doesn't happen when I process the same project in v1.0, it seems like a bug in the current beta. Especially since I just saved the project file as a 1.0 format file, and optimized with 1.0 and it worked fine.

[Alexey Pasumansky]

I have one (disabled) camera with one projection, and a fair number of un-aligned cameras that are blank for projection, but nothing showing 0 projections. Since this behavior doesn't happen when I process the same project in v1.0, it seems like a bug in the current beta. Especially since I just saved the project file as a 1.0 format file, and optimized with 1.0 and it worked fine.

Hello Andy,

We suggest to reset camera alignment for all the cameras that have <5 projections prior to optimization.

Please also provide the console pane output related to the optimization that doesn't do anything in 1.1.