Agisoft PhotoScan 1.3.0 pre-release

[Alexey Pasumansky]

with the command line feature of v1.3 it would be nice to submit the PS model build to a GPU node on a cluster. the issue is that when you submit a job to a cluster you dont know which GPU node will be assigned the job. Since I cant install the license on every GPU node, I get a license error as the result of my run. ( the PS install itself is in a container which is executed on the node). Does anyone know of options to allow for this type of processing?

Hello MManning,

Seems that I've missed your post, can you provide additional details regarding the issue? How you are starting the processing?

[Alexey Pasumansky]

I've run CMD as adminstrator and got the following error:

C:\Program Files\Agisoft\PhotoScan Pro>photoscan --activate and correct code
Error requesting activation
Read error from network (-105)No such file or directory (errno: 2)

Hello ARF,

Seems strange, I still assume that it may be related to the proxy/antivirus/firewall settings, but if you are sure it is not the case, I can suggest to perform offline activation, to do that please execute the following command from CMD:

Code: [Select]

C:\Program Files\Agisoft\PhotoScan Pro>photoscanphotoscan --activate-offline <key> <path/request_file.act>path to the activation request file (*.act) should be accessible with read/write rights, so it's better not to save to the Program Files folder.
Then send the *.act file to support@agisoft.com (don't post it here) and will send you back the license file to be put to PhotoScan installation directory.

[bisenberger]

Shift+scroll wheel in model view changes camera size, nice. 

[bisenberger]

Photoscan stuck at the end of processing Dense Cloud. Check the time and date in console against time in bottom right of screen capture to see how long it has been in this state.

[Alexey Pasumansky]

Hello bisenberger,

It seems that some sub-part of the depth maps estimation has been sent to GPU and was never solved, while all the rest has been calculated by PhotoScan.

I can suggest to update graphic card drivers. If you are using integrated graphics, like Intel HD, it might be reasonable to switch them off.

[mrc]

I am testing 1.3 again after going back to 1.26.

I am scanning fist sized skulls and bones, and my workflow usually involves cleaning up the mounts and supports that I use to photograph the skulls on in the sparse cloud, and then building the dense cloud.

In 1.26, this works well. I get the sparse cloud cleaned up so that just the skull is visible, then create a dense cloud and the skull is reconstructed.

In 1.3, this doesn't work. I get the sparse cloud clean but the dense cloud recreates all of the supports and other objects that I edited out in the sparse cloud. These are significantly more difficult to remove in the sparse cloud (almost impossible...) In 1.3, there are also a lot of odd black objects created around the edges of my model and also in space - this sometimes happens in 1.26 but is many times worse in 1.3

Both processed with depth filtering on mild, high accuracy paired alignment, medium dense cloud.

See screenshots for examples.

[bisenberger]

Hello bisenberger,

It seems that some sub-part of the depth maps estimation has been sent to GPU and was never solved, while all the rest has been calculated by PhotoScan.

I can suggest to update graphic card drivers. If you are using integrated graphics, like Intel HD, it might be reasonable to switch them off.

Graphic card drivers are up to date. Will look into disabling Intel HD although it is not listed under the PhotoScan Preferences GPU tab.

[Srolin]

I am having problems generatin reports, when the process gets to 80% it locks and wont finish. i cannot cancell the process and have to "kill" agisoft in the task manager.

does anuone have that problem to? and how do i fix it?

in advance thank you.

/Soren

[Alexey Pasumansky]

Hello Soren,

I can suggest to update to the build 3149 of the version 1.3.

This issue is the known bug in the build version that you are using.

[Srolin]

Thank you for the quick reply! :-) I will try that

[Alexey Pasumansky]

Hello bisenberger,

Can you then provide the complete log (console output) from the operation that was stuck in the end of depth maps generation stage?

[Alexey Pasumansky]

I am testing 1.3 again after going back to 1.26.

I am scanning fist sized skulls and bones, and my workflow usually involves cleaning up the mounts and supports that I use to photograph the skulls on in the sparse cloud, and then building the dense cloud.

In 1.26, this works well. I get the sparse cloud cleaned up so that just the skull is visible, then create a dense cloud and the skull is reconstructed.

In 1.3, this doesn't work. I get the sparse cloud clean but the dense cloud recreates all of the supports and other objects that I edited out in the sparse cloud. These are significantly more difficult to remove in the sparse cloud (almost impossible...) In 1.3, there are also a lot of odd black objects created around the edges of my model and also in space - this sometimes happens in 1.26 but is many times worse in 1.3

Hello mrc,

The sparse cloud is used only to estimate the scene depth and overlapping image pairs, the dense cloud generation stage then begins with the depth maps generation for the overlapping images.

Looking on the screenshots I assume that in the version 1.2.6 the masks were applied to the images, while in the version 1.3 there are no masks that covers all the background elements.

[mrc]

Hi Alexey,

I am fairly certain the masking did not cover the mounts in the 1.26 reconstruction. Unfortunately I no longer have that reconstruction but I'll reinstall 1.26 and leave it to process overnight and see what happens.

I generally only mask the backgrounds and not the mounts, so it would have been unusual for me to mask them out in this model.

I have since processed another model and there are lots of artefacts (mostly black chunks at margins and in space) in 1.3 that are not there in 1.26.

It is sort of like in 1.26 if there was something in some of the photos and not others it is not reconstructed in the dense cloud, but in 1.3 it is reconstructed. The mounts in the screenshot were only there for about 1/4 of the photos and then the object was flipped over and a new mount was placed.

Thanks

Michael

[mrc]

Hi Alexey,

I ran the reconstruction overnight. The images are not masked, except for the black background. I have attached 2 of the images below.

The result is shown in the screenshots. In 1.26 the dense cloud has none of the stands and props. In 1.3, they are all there along with lots of artefacts.

Please see attached.

Thanks
Michael

[andrey.d]

Hello,

I prepare csv log file for my dataset.
It look like this

Code: [Select]

IMG_2883.JPG,45.3594005,26.712383,431.32,255.82,11.31,-1.95
IMG_2884.JPG,45.3593339,26.7119093,434.13,262.28,2.12,-1.96
IMG_2885.JPG,45.3592974,26.7114327,432.92,267.82,0.48,-2.77And call loadReference() with it. And coordinats added as well.
But by default coordinates in csv file assumed as local coordinates.
How I can set other CS, e.g. EPSG:4326?

Part of code below:

Code: [Select]

crs = PhotoScan.CoordinateSystem("EPSG::3857")
doc = PhotoScan.app.document

doc.open(project_path)
chunk = PhotoScan.app.document.addChunk()

chunk.addPhotos(images_list)
camera = chunk.cameras[0]
chunk.loadReference(out_csv, format=PhotoScan.ReferenceFormatCSV, columns="nxyzabc", delimiter=",", group_delimiters=False, skip_rows=0)
doc.save(project_path, chunks = doc.chunks)
Tanks