Messages

1

Feature Requests / Re: Import ASCII Pts (as laser-like scan)

« on: September 16, 2024, 10:05:08 PM »

Here's the first 7 lines, there is no header:

-0.084851852 0.188710111 0.015139108 -0.853821695 0.037815396 0.519190252
-0.084249358 0.188515195 0.016127928 -0.850836158 0.042110380 0.523740888
-0.083535956 0.188244180 0.017270942 -0.840685785 0.059081525 0.538290560
-0.082543631 0.187915085 0.018839705 -0.833372653 0.069474876 0.548327744
-0.081408911 0.187566130 0.020574075 -0.825069070 0.082025483 0.559046268
-0.081042428 0.187264760 0.021159167 -0.826390386 0.077787474 0.557699025
-0.080428240 0.186837290 0.022114591 -0.817909837 0.087988585 0.568578601

2

Feature Requests / Import ASCII Pts (as laser-like scan)

« on: July 19, 2024, 07:03:50 PM »

We have been using a GOM/Zeiss structured light scanner to supplement some of our photogrammetry. It can export points as ASCII .asc files. I have had some success converting these .asc files to .e57 files via CloudCompare and importing them into Metashape, but it would save a lot of additional work if Metashape could natively import .asc point clouds.

Just a note we are using metashape for making models of objects, not geography/topography. I also know these are structured light scans and not technically laser scans but it is useful particularly when we have objects with heavy occlusions or materials that do not work as well for photogrammetry (flat areas in a solid color with little detail for photogrammetry to latch on to).

If we could import these files directly it would substantially improve our workflow.

3

Bug Reports / Re: Refine Mesh error 1.7.3 12743

« on: July 02, 2024, 08:43:59 PM »

I was curious if this issue with dual matching GPUs had been resolved in any of the Bug fixes since and if so can I remove the CUDA tweak?

4

Bug Reports / Re: Refine Mesh: Kernel Failed: invalid argument (1) at line 99

« on: November 21, 2023, 10:26:01 PM »

I can confirm that the work around was successful.

5

General / Decimated Model not aligned to original

« on: November 17, 2023, 08:42:52 PM »

I produced a model after aligning multiple laser scans to supliment the photogrammetry. I selected the resulting model (with an [R] alignment designation) and decimated it to create a 2nd smaller model. The resulting decimated model is tilted at an angle and not aligned to any other model or point cloud in the project (even in other chunks). This of course makes it impossible to build new textures from the images.

Metashape Professional v. 2.1.0 build 17262 (64 bit)
Windows 10 Pro v. 22H2 OS Build 19045.3693

6

Bug Reports / Re: Refine Mesh: Kernel Failed: invalid argument (1) at line 99

« on: November 08, 2023, 09:02:14 PM »

Original Model: 3,154,562 faces

NVIDIA GeForce RTX 4090 Driver Date: 6/8/2023 Driver Version: 31.0.15.3623

WIndows 10 Pro, Version 22H2 OS Build 19045.3570

7

Bug Reports / Refine Mesh: Kernel Failed: invalid argument (1) at line 99

« on: November 08, 2023, 06:14:48 PM »

I am trying to refine a mesh. I ran it with the settings of Ultra High, 10 (also previously tried iterations, and 0.5 smoothness

The project has 126 aligned images but disabled 40 of them and am trying to only run on 86 images that were cross polarized and not the 40 that were not polarized. All images are 50MP (8272x6200 and 16 bit/channel TIFFS).

Here's a copy and paste from the console:

2023-11-08 09:56:09 Found 2 GPUs in 0.139 sec (CUDA: 0.083 sec, OpenCL: 0.056 sec)
2023-11-08 09:56:09 RefineModel: downscale = 1, iterations = 10, smoothness = 0.5
2023-11-08 09:56:09 Initializing...
2023-11-08 09:56:12 Found 2 GPUs in 0 sec (CUDA: 0 sec, OpenCL: 0 sec)
2023-11-08 09:56:18 Using device: NVIDIA GeForce RTX 4090, 128 compute units, free memory: 22599/24563 MB, compute capability 8.9
2023-11-08 09:56:18   driver/runtime CUDA: 12020/10010
2023-11-08 09:56:18   max work group size 1024
2023-11-08 09:56:18   max work item sizes [1024, 1024, 64]
2023-11-08 09:56:18   got device properties in 0 sec, free memory in 6.584 sec
2023-11-08 09:56:19 Using device: NVIDIA GeForce RTX 4090, 128 compute units, free memory: 22998/24563 MB, compute capability 8.9
2023-11-08 09:56:19   driver/runtime CUDA: 12020/10010
2023-11-08 09:56:19   max work group size 1024
2023-11-08 09:56:19   max work item sizes [1024, 1024, 64]
2023-11-08 09:56:19 Device 'NVIDIA GeForce RTX 4090' has 22209 Mb of free memory
2023-11-08 09:56:19 Device 'NVIDIA GeForce RTX 4090' has 22608 Mb of free memory
2023-11-08 09:56:19 Analyzing model...
2023-11-08 09:56:20 Faces: 3154562, Vertices: 1578983
2023-11-08 09:56:20 Memory required on each device: 2555 Mb + 661 Mb = 3217 Mb
2023-11-08 09:56:20 Using device 'NVIDIA GeForce RTX 4090' in concurrent. (2 times)
2023-11-08 09:56:20 Using device 'NVIDIA GeForce RTX 4090' in concurrent. (2 times)
2023-11-08 09:56:33 Stage #1 out of 2
2023-11-08 09:56:33 Faces: 3154562, Vertices: 1578983
2023-11-08 09:56:33 Memory required on each device: 638 Mb + 661 Mb = 1300 Mb
2023-11-08 09:56:33 Preprocessing model...
2023-11-08 09:56:40 Faces: 3669557, Vertices: 1836481
2023-11-08 09:56:40 Memory required on each device: 2807 Mb + 769 Mb = 3577 Mb
2023-11-08 09:56:41 Loading photos...
2023-11-08 09:57:02 Error: cudaFree(data_): invalid argument (1) at line 108
2023-11-08 09:57:02 Error: cudaFree(data_): invalid argument (1) at line 108
2023-11-08 09:57:02 Error: cudaFree(data_): invalid argument (1) at line 108
2023-11-08 09:57:02 Error: cudaFree(data_): invalid argument (1) at line 108
2023-11-08 09:57:02 loaded photos in 20.686 seconds
2023-11-08 09:57:02 Refining model...
2023-11-08 09:57:12 Iteration #1 out of 10
2023-11-08 09:57:30 Iteration #2 out of 10
2023-11-08 09:57:47 Iteration #3 out of 10
2023-11-08 09:58:04 Iteration #4 out of 10
2023-11-08 09:58:22 Iteration #5 out of 10
2023-11-08 09:58:39 Iteration #6 out of 10
2023-11-08 09:58:56 Iteration #7 out of 10
2023-11-08 09:59:13 Iteration #8 out of 10
2023-11-08 09:59:31 Iteration #9 out of 10
2023-11-08 09:59:48 Iteration #10 out of 10
2023-11-08 10:00:05 Stage #2 out of 2
2023-11-08 10:00:05 Faces: 3669557, Vertices: 1836481
2023-11-08 10:00:05 Memory required on each device: 2555 Mb + 769 Mb = 3325 Mb
2023-11-08 10:00:06 Preprocessing model...
2023-11-08 10:00:25 Faces: 10209967, Vertices: 5106849
2023-11-08 10:00:25 Memory required on each device: 11231 Mb + 2142 Mb = 13373 Mb
2023-11-08 10:00:25 Using device 'NVIDIA GeForce RTX 4090' in concurrent. (1 times)
2023-11-08 10:00:26 Using device 'NVIDIA GeForce RTX 4090' in concurrent. (1 times)
2023-11-08 10:00:26 Loading photos...
2023-11-08 10:00:27 Error: cudaFree(data_): invalid argument (1) at line 108
2023-11-08 10:00:27 Error: cudaFree(data_): invalid argument (1) at line 108
2023-11-08 10:00:27 Finished processing in 257.605 sec (exit code 0)
2023-11-08 10:00:27 Error: Kernel failed: invalid argument (1) at line 99

8

General / Re: What method with photoshop to edit out unwanted item in 9 of 82 photos

« on: April 19, 2023, 07:51:50 PM »

Hi Steve,
Two workflows for this:

If you're working with TIFF files, make the selection, save the selection in photoshop (Select->Save Selection) and choose "New" Channel. It will create a channel called Alpha and that should be all you need. Save the file, go to Agisoft and import masks from file and there's an option to import from alpha channels.

If you're working from things where you cannot save Alpha Channels (JPGs), then you can just make a B&W image, flatten it and save it as a PNG and if you want add _mask to the file name so IMG01234.jpg becomes IMG01234_mask.png. Then in Agisoft there is an option to import masks from files, and you can adjust the file name format so that it looks for the same filename but with _mask.png

Both of these methods will allow you to save only some images with masks and when Agisoft imports the masks, if none are found for a given file no mask will be created for that image.

9

Feature Requests / Re: Instant NERF?

« on: April 17, 2023, 08:57:47 PM »

Photogrammetry and NeRFs are two different things but there is some overlap, particularly in the capture method. Also NeRF does require having camera locations, and they're using COLMAP SfM as a separate script to determine the camera locations. If we've done photogrammetry on a scene or object and have done the work to optimize the camera locations, it would be nice to just be able to export it in a format that could be fed into creating a NeRF. There are some use cases where NeRFs might be desirable. Maybe the material properties don't work well for photogrammetry but want to see if rendering a video from a NeRF might give a useable representation.

While I don't expect Agisoft to build NeRFs, a simple to implement feature that would be useful would be if we could export the cameras in the JSON format that NVidia's InstantNeRF is using so that we could take the images and try them in NeRF.