[Open3D_Cuda] Adjust depth truncation in TSDF integration


I’m using the Open3D CUDA version to do TSDF integration. The default setting is that only points with depth between [0.1m,3.5m] are considered. I’d like to extend it to up [0.1m, 10.0m] and I did two modifications.

  1. In “TouchSubvolume”, it truncates depth >3.5m. I adjust it to truncate depth >10.0m.
  2. In constructing RGBDImageCuda, I set depth_truncate = 10.0m.

However, the program frequently crashed during integration with the following message,

Processing … /tum/rgbd_dataset_freiburg3_long_office_household/depth/1341847983.066757.png
[Open3D INFO] Active subvolumes in volume: 4113
CUDA error at /home/lch/Code_ws/Open3D_Dong/Open3D/src/Cuda/Integration/ScalableTSDFVolumeCudaKernel.cuh:113 code=700(cudaErrorIllegalAddress) “cudaDeviceSynchronize()”

I believe it is related to the memory allocation in CUDA. But I’m not sure the exact reason.

If anyone can help, thanks a lot.