Agisoft PhotoScan 1.3.0 pre-release

[Alexey Pasumansky]

Hello Airgo,

Ok, thanks for the information. Actually, SPIR support line should be even displayed in the console for NVIDIA cards, as for them PhotoScan is using CUDA now.

But as for the spherical data processing, we have already found the issue and will also include the fix in the next update.

[RHenriques]

Hi Alexey

It seems that Airgo problem with the license is the same that I'm facing. Do you have any successful story of version 1.3 beta installation in MacOS Sierra?
Cheers

[MManning]

Hello
I am trying to duplicate the processing of images in batch mode that my researcher does via the UI. She follows this path:

- load images
- build dense cloud
- build mesh (buildModel)
- build texture
- build orthomosaic[/right][/right]

However when I hit the buildTexture step in my batch process I get this error:
RuntimeError: Missing uv coordinates

In the example in the Photoscan 1.3 docs it show that buildUV step is before buildTexture. Is this a requirement for buildTexture? My researcher says that she does not perform a buildUV-type step in the UI that she is aware of.

Any ideas on what option I might be missing on my build steps?

Thanks

[mattcalavera]

I'm trying to compare PS version 1.2.6 and 1.3.0.

I Use a workspace of 416 photos from inspire 1|X5

in version 1.2.6 all 416 photos correctly aligned

in version 1.3.0 (with the same alignment settings) only 211 photos aligned, and nothing change also if I use maximum alignment settings (Highest accuracy, etc)


Why?

Thanks,
Matt

[Alexey Pasumansky]

Hello Matt,

Can you provide the processing logs from both tries and also the report files generated from both versions to support@agisoft.com?

[Alexey Pasumansky]

Hello
I am trying to duplicate the processing of images in batch mode that my researcher does via the UI. She follows this path:

- load images
- build dense cloud
- build mesh (buildModel)
- build texture
- build orthomosaic[/right][/right]

However when I hit the buildTexture step in my batch process I get this error:
RuntimeError: Missing uv coordinates

In the example in the Photoscan 1.3 docs it show that buildUV step is before buildTexture. Is this a requirement for buildTexture? My researcher says that she does not perform a buildUV-type step in the UI that she is aware of.

Any ideas on what option I might be missing on my build steps?

Thanks

Hello MManning,

Do you mean Python scripting?

If so, then the GUI Align Photos option consist of two Python operations: .matchPhotos() and .alignCameras(). Similar is with the GUI Build Texture option - in Python it is split to .buildUV() and .buildTexture().

[MManning]

Hi Alexey
thanks for the response. When I add in the buildUV step before the buildTexture:

doc = PhotoScan.app.document
doc.open("...")
chunk = doc.chunk
chunk.buildUV(mapping=PhotoScan.GenericMapping)
chunk.buildTexture( color_correction=True )
chunk.buildOrthomosaic( surface=PhotoScan.DataSource.ModelData )
doc.save()


I get an error :

'blending textures... *********************************** frames quality estimated in 1.23499 sec', '*********************************** done in 13.5811 sec', 'postprocessing atlas... done in 0.011684 sec', 'BuildOrthomosaic: projection type = Planar, blending mode = Mesh, surface = Mosaic, resolution = 0', 'Please save project in PSX format before processing', 'Traceback (most recent call last):', '  File "/global/home/users/mmanning/actest/commandscript.sh", line 10, in <module>', '    chunk.buildOrthomosaic( surface=PhotoScan.DataSource.ModelData ) ', 'RuntimeError: Empty frame path'

Im not sure what to do with this, I see only one reference to a PSX in the 1.3 ref documentation. Do I need to export the model into a PSX then run build texture? Is the "Empty frame path" message related or something different?

Thank you for your assistance, still learning Photoscan.

Maurice

[Alexey Pasumansky]

Hello MManning,

According to the log the texture is built. But you need to save the project in PSX format to build orthomosaic (it's not connected anyhow to the texture).

[mattcalavera]

Hello Matt,

Can you provide the processing logs from both tries and also the report files generated from both versions to support@agisoft.com?

Alexey, mail sent....

[MManning]

Hello MManning,

According to the log the texture is built. But you need to save the project in PSX format to build orthomosaic (it's not connected anyhow to the texture).

ok,  after reading the docs I understand the difference between a psz and psx a little bit better however Im not clear, when using the python sdk, how to save to each format.

- It seems from the photoscan python api document that doc.save() seems to create a .psz by default, is that true?

- if I just name the file with the project extension (  doc.save("project.psx"  ) will photoscan save to the correct (project vs archive) format?

- is Model.saveTexture("project.psx") preferred way to save saving before building the orthomosaic?

- is it true that archive (psz) format is the best format for holding the full set of information (images, settings, etc)  for long term?

Again, thanks for your help and patience.
Maurice

[Alexey Pasumansky]

Hello Maurice,

If you pass to doc.save() function the path to the .psx project, it will be saved in psx format. For example: doc.save("D://projects/new_project.psx")
So the function is checking the extension in the provided path to define the save format.

Model.saveTexture("project.psx") is meaningless, you need to put doc.save(some_psx_path) line before generating the orthomosaic.

PSZ - is a single archive file, while PSX is the folder structure. PSZ cannot contain orthomosaic, individually orthorectified images, DEM and tiled model. Also when PSZ file is opened it is fully loaded to the memory, whereas when working with PSX format PhotoScan will load only required project element.

[mattcalavera]

Hello Matt,

Can you provide the processing logs from both tries and also the report files generated from both versions to support@agisoft.com?

Alexey, mail sent....

Hello Alexey, this night I made another test, with version 1.3.0 disabling GPU, and yes without GPU it align correctly all the photos, so I think the problem is with the align algoritm that use the GPU..... I sent a mail with the three reports and logs ( 1.2.6 with GPU, 1.3.0 with GPU, and 1.3.0 Without GPU)....

Let me know...

thanks,

Matt

[Alexey Pasumansky]

Hello Matt,

Thanks, we are checking it.

[Alexey Pasumansky]

Hello Matt,

We have requested the sample dataset, but it seems that it's no longer needed, as we were able to reproduce the problem on multi-GPU setup.
Could you please also check the alignment on your system with only one of two GPU devices one, just to be sure that it works properly in your case?

[Alexey Pasumansky]

It seems that Airgo problem with the license is the same that I'm facing. Do you have any successful story of version 1.3 beta installation in MacOS Sierra?

Hello RHenriques,

It was a different issue, that was related to the hardware change.

We've tried to reproduce your problem on MacOS Sierra (clean install), but it worked as expected for current pre-release build 3520 - both for Applications and random installation directory, so I assume that there might be some settings in OS that moves the app bundle to AppTranslocation folder.

Aren't you running PhotoScan from .dmg? You need to copy PhotoScan Pro application bundle from the dmg somewhere before running it.