GitLab

If there are less than half of model_cfg.NUM_KEYPOINTS it will lead to 
RuntimeError: unsupported operation: some elements of the input tensor and the written-to tensor refer to a single memory location. 
Please clone() the tensor before performing the operation.

fixbug: multi_classes_nms and waymo dataset with old version splits

fixbug: modify input args with return for box utils

fixbug: AssertionError features.is_contiguous() in pointnet2_utils.py

fixbug: error of sample_points when num_point < far_idx_choice

caused by `features` in pointnet2_backbone.py:78, which is possible to be a non-contiguous tensor, being directly passed into SA_modules without calling .contiguous()

In data_processor.py sample_points, when num_points < far_idx_choice, a negative value is passed into np.random.choice, which cause a ValueError.
In fact the variable `near_idxs_choice` return by the deleted line is never used.

* bugfixed: add num_points_in_gt to the waymo_infos for evaluation

* add clarification for Waymo evaluation

* add kitti_utils.py for evaluating with KITTI metrics

* bugfixed: ignore empty boxes in visualization

* add contiguous() after transpose() at pointrcnn_head.py

* automatically find unused tcp port when mgpu train or test

* fixbug: set tcp_port before set python args

bugfixed: update slurm_train.sh

Release OpenPCDet v0.3.0