BADet: Boundary-Aware 3D Object Detection from Point Clouds (Pattern Recognition 2022)

As of Apr. 17th, 2021, 1^st place in KITTI BEV detection leaderboard and on par performance on KITTI 3D detection leaderboard. The detector can run at 7.1 FPS.

Authors: Rui Qian, Xin Lai, Xirong Li

[arXiv] [elsevier]

Citation

If you find this code useful in your research, please consider citing our work:

@InProceedings{qian2022pr,
author = {Rui Qian and Xin Lai and Xirong Li},
title = {BADet: Boundary-Aware 3D Object Detection from Point Clouds},
booktitle = {Pattern Recognition (PR)},
month = {January},
year = {2022}
}
@misc{qian20213d,
title={3D Object Detection for Autonomous Driving: A Survey}, 
author={Rui Qian and Xin Lai and Xirong Li},
year={2021},
eprint={2106.10823},
archivePrefix={arXiv},
primaryClass={cs.CV}
}

Updates

2021-03-17: The performance (using 40 recall poisitions) on test set is as follows:

Car [email protected], 0.70, 0.70:
bbox AP:98.75, 95.61, 90.64
bev  AP:95.23, 91.32, 86.48 
3d   AP:89.28, 81.61, 76.58 
aos  AP:98.65, 95.34, 90.28 

Introduction

Currently, existing state-of-the-art 3D object detectors are in two-stage paradigm. These methods typically comprise two steps: 1) Utilize a region proposal network to propose a handful of high-quality proposals in a bottom-up fashion. 2) Resize and pool the semantic features from the proposed regions to summarize RoI-wise representations for further refinement. Note that these RoI-wise representations in step 2) are considered individually as uncorrelated entries when fed to following detection headers. Nevertheless, we observe these proposals generated by step 1) offset from ground truth somehow, emerging in local neighborhood densely with an underlying probability. Challenges arise in the case where a proposal largely forsakes its boundary information due to coordinate offset while existing networks lack corresponding information compensation mechanism. In this paper, we propose $BADet$ for 3D object detection from point clouds. Specifically, instead of refining each proposal independently as previous works do, we represent each proposal as a node for graph construction within a given cut-off threshold, associating proposals in the form of local neighborhood graph, with boundary correlations of an object being explicitly exploited. Besides, we devise a lightweight Region Feature Aggregation Module to fully exploit voxel-wise, pixel-wise, and point-wise features with expanding receptive fields for more informative RoI-wise representations. We validate BADet both on widely used KITTI Dataset and highly challenging nuScenes Dataset. As of Apr. 17th, 2021, our BADet achieves on par performance on KITTI 3D detection leaderboard and ranks $1^{st}$ on $Moderate$ difficulty of $Car$ category on KITTI BEV detection leaderboard. The source code is available at https://github.com/rui-qian/BADet.

Dependencies

• python3.5+
• pytorch (tested on 1.1.0)
• opencv
• shapely
• mayavi
• spconv (v1.0)

Installation

1. Clone this repository.
2. Compile C++/CUDA modules in mmdet/ops by running the following command at each directory, e.g.

$ cd mmdet/ops/points_op
$ python3 setup.py build_ext --inplace

3. Setup following Environment variables, you may add them to ~/.bashrc:

export NUMBAPRO_CUDA_DRIVER=/usr/lib/x86_64-linux-gnu/libcuda.so
export NUMBAPRO_NVVM=/usr/local/cuda/nvvm/lib64/libnvvm.so
export NUMBAPRO_LIBDEVICE=/usr/local/cuda/nvvm/libdevice
export LD_LIBRARY_PATH=/home/qianrui/anaconda3/lib/python3.7/site-packages/spconv;

Data Preparation

1. Download the 3D KITTI detection dataset from here. Data to download include:

   • Velodyne point clouds (29 GB): input data to VoxelNet
   • Training labels of object data set (5 MB): input label to VoxelNet
   • Camera calibration matrices of object data set (16 MB): for visualization of predictions
   • Left color images of object data set (12 GB): for visualization of predictions

2. Create cropped point cloud and sample pool for data augmentation, please refer to SECOND.

3. Split the training set into training and validation set according to the protocol here.

4. You could run the following command to prepare Data:

$ python3 tools/create_data.py

qr@Misaki:~/qianrui/kitti$ tree -L 1
data_root = '/home/qr/qianrui/kitti/'
├── gt_database
├── ImageSets
├── kitti_dbinfos_train.pkl
├── kitti_dbinfos_trainval.pkl
├── kitti_infos_test.pkl
├── kitti_infos_train.pkl
├── kitti_infos_trainval.pkl
├── kitti_infos_val.pkl
├── train.txt
├── trainval.txt
├── val.txt
├── test.txt
├── training   <-- training data
|       ├── image_2
|       ├── label_2
|       ├── velodyne
|       └── velodyne_reduced
└── testing  <--- testing data
|       ├── image_2
|       ├── label_2
|       ├── velodyne
|       └── velodyne_reduced

Pretrained Model

You can download the pretrained model [Model][Archive], which is trained on the train split (3712 samples) and evaluated on the val split (3769 samples) and test split (7518 samples). The performance (using 11 recall poisitions) on validation set is as follows:

[40, 1600, 1408]
[>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 3769/3769, 7.1 task/s, elapsed: 533s, ETA:     0s
Car [email protected], 0.70, 0.70:
bbox AP:98.27, 90.22, 89.66
bev  AP:90.59, 88.85, 88.09
3d   AP:90.06, 85.75, 78.98
aos  AP:98.18, 89.98, 89.25
Car [email protected], 0.50, 0.50:
bbox AP:98.27, 90.22, 89.66
bev  AP:98.31, 90.21, 89.73
3d   AP:98.20, 90.11, 89.61
aos  AP:98.18, 89.98, 89.25

Quick demo

You could run the following command to evaluate the pretrained model:

cd mmdet/tools
# vim ../configs/car_cfg.py(modify score_thr=0.4, score_thr=0.3 for val split and test split respectively.)
python3 test.py ../configs/car_cfg.py ../saved_model_vehicle/epoch_50.pth

Training

To train the BADet with single GPU, run the following command:

cd mmdet/tools
python3 train.py ../configs/car_cfg.py

Inference

To evaluate the model, run the following command:

cd mmdet/tools
python3 test.py ../configs/car_cfg.py ../saved_model_vehicle/latest.pth

Acknowledgement

The code is devloped based on mmdetection, some part of codes are borrowed from SA-SSD, SECOND, and PointRCNN.

• mmdetection
• mmcv
• second.pytorch
• PointRCNN
• SA-SSD

Contact

If you have questions, you can contact [email protected].