Spatial Pruned Sparse Convolution for Efficient 3D Object Detection

Spatial Pruned Sparse Convolution for Efficient 3D Object Detection (NeurIPS 2022)
By Jianhui Liu, Yukang Chen, Xiaoqing Ye, Zhuotao Tian, Xiao Tan, and Xiaojuan Qi.

Introduction

3D scenes are dominated by a large number of background points, which is redundant for the detection task that mainly needs to focus on foreground objects. In this paper, we analyze major components of existing sparse 3D CNNs and find that 3D CNNs ignore the redundancy of data and further amplify it in the down-sampling process, which brings a huge amount of extra and unnecessary computational overhead. Inspired by this, we propose a new convolution operator named spatial pruned sparse convolution (SPS-Conv), which includes two variants, spatial pruned submanifold sparse convolution (SPSS-Conv) and spatial pruned regular sparse convolution (SPRS-Conv), both of which are based on the idea of dynamically determining crucial areas for redundancy reduction. We validate that the magnitude can serve as important cues to determine crucial areas which get rid of the extra computations of learning-based methods. The proposed modules can easily be incorporated into existing sparse 3D CNNs without extra architectural modifications. Extensive experiments on the KITTI, Waymo and nuScenes datasets demonstrate that our method can achieve more than 50% reduction in GFLOPs without compromising the performance.

Experimental results

nuScenes dataset

	mAP	NDS	download
CenterPoint + SPSS ratio0.5 + SPRS ratio0.5	58.68	66.26	Google
CenterPoint + SPSS ratio0.3 + SPRS ratio0.5	59.09	66.63	Google

Getting Started

Installation

a. Clone this repository

https://github.com/CVMI-Lab/SPS-Conv.git && cd SPS-Conv

b. Install the environment

Following the install documents for OpenPCdet

*spconv 2.x is highly recommended instead of spconv 1.x version.

c. Prepare the datasets.

Download and organize the official KITTI and Waymo and nuScenes following the document in OpenPCdet.

Training

For KITTI

cd tools 
bash scripts/dist_train.sh ${NUM_GPUS} --cfg_file /cfgs/kitti_models/voxel_rcnn_car_spss_ratio0.5_sprs_ratio0.5.yaml

For nuScenes

cd tools 
bash scripts/dist_train.sh ${NUM_GPUS} --cfg_file cfgs/nuscenes_models/cbgs_voxel0075_res3d_centerpoint_spss_ratio0.3_sprs_ratio0.5.yaml

For Waymo

cd tools 
bash scripts/dist_train.sh ${NUM_GPUS} --cfg_file cfgs/waymo_models/centerpoint_spss_ratio0.3_sprs_ratio0.5.yaml

Evaluation

We provide the trained weight file so you can just run with that. You can also use the model you trained.

NUM_GPUS=4
cd tools 
bash scripts/dist_test.sh ${NUM_GPUS} --cfg_file ${PATH_TO_CONFIG_FILE} --ckpt ${PATH_TO_CHECKPOINT}

TODO List

- CUDA code of SPS-Conv
- KITTI and Waymo pre-trained weights (will be released within a week)

Citation

If you find this project useful in your research, please consider citing:

@article{liu2022spatial,
  title={Spatial Pruned Sparse Convolution for Efficient 3D Object Detection},
  author={Liu, Jianhui and Chen, Yukang and Ye, Xiaoqing and Tian, Zhuotao and Tan, Xiao and Qi, Xiaojuan},
  journal={arXiv preprint arXiv:2209.14201},
  year={2022}
}

Acknowledgement

This work is built upon the OpenPCDet. Please refer to the official github repositories, OpenPCDet for more information.

yukang2017 / SPS-Conv