PCSeqLearning

This is the code repository for the paper:

Representation Learning for Object Detection from Unlabeled Point Cloud Sequences.
Xiangru Huang, Yue Wang, Vitor Guizilini, Rares Ambrus, Adrien Gaidon and Justin Solomon.
Conference on Robotic Learning (CoRL) 2022.

The code is adapted from OpenPCDet. The code will be updated by September 5th to include full functionality. Currently, it can extract object cluster sequences from LiDAR point cloud sequences for Waymo Open Dataset, which is the major algorithmic part in the paper.

System Requirements

The code has been tested with the following (major) environment dependencies:

waymo-open-dataset-tf-2-5-0==1.4.3
torch==1.13.1+cu117

Installation

Please check Install.md for instructions of setting up this repo.

Data preparation

Please follow the instructions here for preparing Waymo Open Dataset.

Demo

pip install polyscope
bash scripts/dist_train_multi.sh 0 cfgs/waymo_models/PCsequence/registration/cluster_tracking_TLS_multiradius_every8.yaml cfgs/dataset_configs/waymo/PCsequence/registration/all_sequence.yaml cfgs/optimizers/registration.yaml --vis_cfg_file cfgs/visualizers/waymo/PCsequence/registration/voxel_visualizer.yaml

The visualizer will show the extracted object clusters given an input Waymo point cloud sequence.