Forum

Recent Posts

Pages: [1] 2 3 ... 10
1
I'm working with a supercomputing cluster to develop a set of python scripts to run computationally-intensive processes remotely on an HPC node. We've sorted out the photo alignment, dense cloud/depthmaps, and mesh steps using the current Agisoft Python manual, but we're having trouble with the texture script.

So far, we've been unable to run a command for texture alone, even on a chunk for which all other steps have already been completed. The only working version of the script is the same as the model in the guide:

chunk.alignCameras()
chunk.buildDepthMaps(downscale=2, filter_mode=ms.FilterMode.ModerateFiltering)
chunk.buildDenseCloud( point_colors=True, point_confidence=True )
chunk.buildModel()
chunk.buildUV()
chunk.buildTexture()

This seems not to overwrite the optimization work we've done manually after the initial alignment, but it does overwrite any dense-cloud editing that we've performed manually after the dense-cloud generation stage. It also takes a much longer time than necessary, since it's re-running the earlier steps, which is a problem when we have limited HPC cycles.

The models involved are often too large to let us easily run the texture process through the GUI on a local machine. Does anyone have a script that will only carry out the texture steps (buildUV, buildTexture) on a chunk for which the other steps are already complete? We're not sure where the problem is arising. Thanks!

2
Other Languages / Re: Parrot Sequoia + Agisoft Metashepe
« Last post by Alexey Pasumansky on April 22, 2021, 07:30:58 PM »
Для тех, кто может испытывать похожую проблему: при добавлении снимков с Parrot Sequoia камеры следует выбирать только TIFF файлы, соответствующие мультиспектральным сенсорам. Одновременно с ними не нужно выбирать JPG файлы, соответствующие RGB сенсору.
3
General / Re: Alignment time v1.7.3 vs v1.6.6
« Last post by Paulo on April 22, 2021, 02:34:36 PM »
Ok,

then I can understand the different times and quantity of Tpts determined using same parameters for both versions...Thank you for the clarification!
4
General / Re: Alignment time v1.7.3 vs v1.6.6
« Last post by Alexey Pasumansky on April 22, 2021, 02:12:31 PM »
Hello Paul,

In 1.7 there were changes both in image matching and alignment that should improve the stability of alignment, especially in previously unstable cases.
5
Bug Reports / Processed thumbnails load too slowly
« Last post by cbnewham on April 22, 2021, 10:37:13 AM »
Thumbnails load incredibly slowly once the depth maps have been generated.

What's worse is that only a small number are loaded - if you scroll down, it forces more to load and the earlier ones disappear. It doesn't behave like this when the photos are initially loaded - everything is loaded super fast and always visible. It's been like this on every version up to the one I'm using (1.7.0) and I've always assumed it was a bug.

If this is not a bug, please provide a way to switch this behaviour off (so that only the thumbnail is loaded, and not any other background information). With many hundreds or thousands of photos it makes the thumbnails unusable.
6
General / Re: Ruler tool on 2D photo
« Last post by Paulo on April 22, 2021, 02:01:10 AM »
Yes Kaaiinn,

that is correct though difference between the 2 measurements (using sparse cloud or dense cloud) seems to be small....

In example, i have 2 markers (1 and 2) on a photo (Drohne01634) and I print out their image coordinates, world coordinates and distance between the 2 using first point_cloud (sparse cloud) as surface and then dense_cloud. Difference is at cm or mm level....

For 2 markers set further apart on same photo, the difference in distance between using sparse or dense cloud is still small (2cm). See 2nd attachment...

Another example where distance calculated using 3dModel as surface vs point cloud shows much more discrepancy around 55 cm...3rd attachmant
7
General / Re: Ruler tool on 2D photo
« Last post by kaaiinn on April 22, 2021, 12:07:55 AM »
Thank you very much for replying!


Just so I understand what you are saying -

When viewing the image and using the ruler function the program projects this point onto the sparse point cloud? (I have not yet made the dense point cloud) I am pretty amazed at the apparent accuracy of this as I have the ability to measure small objects - when I am pretty sure there would not be many tie points around it.

Apologies, I am not familiar with the pickPoint function API
8
General / Re: Alignment time v1.7.3 vs v1.6.6
« Last post by Paulo on April 21, 2021, 11:02:28 PM »
Thanks Alexey,

so you are still getting relatively higher alignment times for 1.7.3 vs 1.6.6. Is that normal? And if algorythm has not changed, why does 1.7.3 find more initial tpts than 1.6.6?

In my experience, using low performance graphics card as GTX 670 MX, processing times are lower when enabling CPU while GPU processing. Is that your experience on your side?

Again for your insight on this issue...
9
General / Re: Ruler tool on 2D photo
« Last post by Paulo on April 21, 2021, 10:55:21 PM »
ni kaiin,

I guess the ruler function on photo works thi way:

- for each ruler click or point (2D) on photo, the program  projects this point onto the model (using pickPoint function in API) and gets its 3d coordinates in real world or local coordinates;
- the distance is calculated   from these 2 3d point coordinates.

Now, if there is no model, then it uses dense cloud. If no dense cloud then it uses the sparse cloud. Obviously Model or dense cloud would give more precise results than sparse cloud.
10
General / Re: Point Dense Cloud "Zero resolution" error
« Last post by Alexey Pasumansky on April 21, 2021, 10:46:51 PM »
Hello BitSaginaw,

If possible, please provide the screenshot of the tie point cloud with the estimated camera location from Metashape Model view.

Also please check, if you see any adequate depth maps preview images in the Photos Pane with "show depth maps" button pressed - you should see some RGB pictures.
Pages: [1] 2 3 ... 10