Closed chrockey closed 3 years ago
Yes, the sparse quantize now do not support label collision in v0.5
From v0.5, I recommend not quantizing the point cloud during the data loading and use .slice()
on a sparse tensor to recover the original point cloud like the example https://github.com/NVIDIA/MinkowskiEngine/blob/master/examples/indoor.py#L152.
Describe the bug I'm not sure it is intended that 'sparse_quantize' function does not support label collision now. https://github.com/NVIDIA/MinkowskiEngine/blob/7c3b87ffaf2fa4c5ee452b513b0f781d988d1bfa/MinkowskiEngine/utils/quantization.py#L269
To Reproduce Use sparse_quantize function with label collision.
Expected behavior Return ignore_label when there is label collision in a voxel.