GPU point cloud

Is there a specific command you need to add to the script file for the ODM to use the GPU in the point cloud stages, or is the GPU supposed to automatically detect itself? Recent tests indicate that it appears to be failing on the reconstruction when attempting to use GPU.


If everything is set up correctly, ODM will detect the GPU and use it for whatever stage applicable. Though, just to mention it, you will need a NVIDIA card with CUDA capability for using with ODM.

What do you mean? Failing on the reconstruction?
If everything goes right you should see:

[INFO]    nvidia-smi detected
[INFO]    Using GPU for extracting SIFT features

That is a line in the console output or console.txt if you download it.

Because of a limitation with CUDA or the CUDA implementation you could also see:

[INFO]    nvidia-smi detected
[WARNING] Image size (5472x3648px) would not fit in GPU memory, falling back to CPU

But that ONLY concerns feature extraction. Point cloud computation will still happen on the GPU, as long as you see a line with nvidia-smi detected in your console output.

Or did the process fail / not finish when using the GPU?

1 Like

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.