heiwang1997
commented
1 year ago Hey you could specify the device, e.g. cuda:1 or cuda:2 in order to use multiple GPUs. You have to manually split your input into multiple chunks and send them to Reconstructors residing on different devices, in order to leverage multiple GPUs.
Regarding your IndexError, could you show me your input point cloud? This usually means that it is too small so that no surface could be extracted (but could be also due to our bug).
hello, can I use multiple PGUs during inference? when I reconstruct the point cloud in chunks, my input cloud is not empty, but the code reports an error : svh = SparseFeatureHierarchy.joined([f.svh for f in fields], transforms) File "/home/panosvrach/PythonEnv/nksr/lib/python3.8/site-packages/nksr/svh.py", line 317, in joined ref_svh = svhs[0] IndexError: list index out of range