Open AndyYuan96 opened 2 years ago
I use dynamic voxelization to train nus model, I find sometime's voxelize will generate negative index, which make index operation out of range, even if I already use PointsRangeFIlter. In this line : https://github.com/open-mmlab/mmdetection3d/blob/eb5a5a2d166d2d60dfbdffe63b925cb37a6541e3/mmdet3d/models/detectors/mvx_faster_rcnn.py#L39 res_coors = self.pts_voxel_layer(res)
I will fix it asap.
Do you have any updates/more info on this? I run into the same issue from time to time.
I use dynamic voxelization to train nus model, I find sometime's voxelize will generate negative index, which make index operation out of range, even if I already use PointsRangeFIlter. In this line : https://github.com/open-mmlab/mmdetection3d/blob/eb5a5a2d166d2d60dfbdffe63b925cb37a6541e3/mmdet3d/models/detectors/mvx_faster_rcnn.py#L39 res_coors = self.pts_voxel_layer(res)