An official implementation of AAAI2023 paper "LWSIS: LiDAR-guided Weakly Supervised Instance Segmentation for Autonomous Driving"

Here we explain different annotations used in the exp. 'box' means only using the 2D bounding box annotation for each instance, 'point' means using a specific number of points with human annotation indicating the background/foreground, '3dbox' means using the 3D bounding box annotations for each instance and 'pc' means the original point cloud.

First install Detectron2 following the official guide: INSTALL.md.

Please use Detectron2 with commit id 9eb4831 if you have any issues related to Detectron2.

Then build LWSIS with:

git clone [email protected]:Serenos/LWSIS.git
cd LWSIS
python setup.py build develop

• Download the nuscenes origin datasets to ${HOME}/datasets/. The folder structure shall be like this:

  • nuscenes
    • annotations
    • lidarseg
    • maps
    • samples
    • sweeps
    • v1.0-trainval
    • v1.0-mini

• Download nuInsSeg3d_train(访问码：4aml), nuInsSeg3d_val(访问码：luw8) and put it into the nuscenes/annotations folder.

• Training

  bash tools/train.sh configs/BoxInst/MS_R_50_1x_nuscenes.yaml Boxinst_LWSIS 000
  

• Evaluation

  bash tools/test.sh configs/BoxInst/MS_R_50_1x_nuscenes.yaml output/Boxinst_LWSIS/000/model_final.pth 
  

We supplement instance mask annotation for nuScenes dataset. For more detail, please follow the nuinsseg-devkit.

The authors are grateful to School of Computer Science, Beijing Institute of Technology, Shanghai AI Laboratory, Inceptio, 4SKL-IOTSC, CIS, University of Macau.

The code is based on Adlaidet.