ChaimZhu

I can not reproduce the `NaN` result using your provided example. It calculates the square dist correctly.

> Great, `torch.cdist` would be a better solution for this situation, contributions ( PR) are welcome if you have time after checking the performance influence of this modification.

The CPU op may need to be rewritten. For now, it's directly moved from original mmdet3d repo, which is never used before.

If you want to run the model on test dataset, you need to change `kitti_infos_val.pkl` to `kitti_infos_test.pkl`, then just follow the instructions in `Testing and make a submission` parf of...

yes, you need to change it to `testing` as well, `split` means split of input data. KITTI dataset is divided into `testing` and `training` part after data preparation, you can...

`RuntimeError: mat1 dim 1 must match mat2 dim 0.` means that the rotation matrix dimension does not match, maybe like a situation: [3x3] x [4x3]. Besides, can you provide the...

Did you train SECOND on your own dataset or on kitti?

Sorry for the late reply, for now, we provide the `browse_dataset.py` script which can visualize the augmented data in the training pipeline, more details can be seen in https://github.com/open-mmlab/mmdetection3d/blob/master/tools/misc/browse_dataset.py. Because...

It seems that you didn't compile mmdet3d succesfully. Did you run `python setup.py develop` in the root of MMDetection3d?

Please refer to https://github.com/open-mmlab/mmdetection3d/issues/1008