resume batch.

[WickedAndy]

having a issue I have a scene that has 100's of chunks... photoscan gets about 30 - 40 chunks in usually before experiencing a crash, I have it saving after every step, it would be great to be able to resume, having an option to "skip existing, geometry, textures ect.. " would be handy.

[Alexey Pasumansky]

Hello Andy,

Could you please describe the conditions when PhotoScan crashes after 30-40 chunks processing?
What process is performed (align photos, build geometry) and what OS are you using? How many photos are there in chunks?

Maybe you've noticed something special that mostly leads to crash. It would we very useful information for us to fix this problem.

[WickedAndy]

here's my process.
I am currently testing using photoscan to do facial capture, I have builta small test rig currently using only 2 cameras (eventually 10 or more).

I bring in each pair of images into a new chunk of the sequence that I am processing, I then import cameras from a previous exported frame, so each alignment happens in the same world space.

I've only been testing with about 100 chunks at at time currently, since it's a bit labor intensive. altho I did create a helper utiltiy that aid's in setting up the initial .psz scene. (just loading images into chunks)

Alignment is not being run in batch since I have to import cameras for every chunk. crash happens when building geometry, it happens in batch and if I manually build, somewhere between 20 or 30 chunks further once I restart, it seems to crash. batch texture generation appears to work properly. 

I am running windows 7, on a i7-2600k cpu, w/Nvidia GTX580 driver ver 275.33

:UPDATE:
I just ran a test disabling OpenCL in preferences and it appears that the crashes do not occur, it did take quite
a lot longer to process but it did successfully finish.

I will send you my scene and image files that are causing the crash.


thanks and it is an truly awesome program....

[IanJohnson]

Do each of your chunks only have two photos?  Are you doing it this way so that movement doesn't affect the creation of the geometry? 

[WickedAndy]

yes for the moment I only have 2 images per chunk, as the cameras are in a fixed configuration, the cameras have sync'd shutters to capture at exactly the same time.

 I will expand the system shortly, did some testing, seems about 6 cameras should give me pretty decent coverage of about 180 deg of a persons face, I also have the capability to capture at over 100 frames per second, as to be able to temporally average out noise in the capture data, since stereo reconstruction tends to be noisy for this type of application.

 I intend to do a write up and post some examples once I am further in my testing.