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

Models

Model	Backbone	Annotations	Lr_schedule	Mask_AP	Download
BoxInst	R-50	box	1x	33.65	link(访问码：pmw0)
BoxInst	R-101	box	1x	34.39	link
PointSup	R-50	box+point	1x	43.80	link
PointSup	R-101	box+point	1x	44.72	link
LWSIS+BoxInst	R-50	3dbox+pc	1x	35.65	link(访问码：hy6a)
LWSIS+BoxInst	R-101	3dbox+pc	1x	36,22	link
LWSIS+PointSup	R-50	3dbox+pc	1x	45.46	link
LWSIS+PointSup	R-101	3dbox+pc	1x	46.17	link

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.

Install

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 git@github.com:Serenos/LWSIS.git
cd LWSIS
python setup.py build develop

Quick Start

• 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 
  

nuInsSeg Dataset and devkit

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

Acknowledgements

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.