Semantic Occupancy Prediction with Dual Range-Voxel Representation

Environment

1. Create the conda environment

conda create -n occ python=3.7
conda activate occ

2. Install the following dependencies

pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu116
pip install torch-scatter -f https://data.pyg.org/whl/torch-1.12.0+cu116.html
pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu116/torch1.12.1/index.html
pip install spconv-cu116 tensorboard numba nuscenes-devkit

Prepare Dataset

1. SemanticKITTI

   • Download SemanticKITTI from SemanticKITTI
   • Download SSC labels from SemKitti-SSC
   • Modify utils/preprocess.py and execute it to preprocess SSC labels.

2. SemanticPOSS

   • Download SemanticPOSS from SemanticPOSS
   • Download SSC labels from SemPoss-SSC

3. nuScenes-Occupancy

   • Download nuScenes from nuScenes
   • Download OCC labels from nuScenes-Occupancy
   • Download the generated train info file from Google Drive.
   • Download the generated val info file from Google Drive.

4. Occ3D-nuScenes

   • Download nuScenes from nuScenes
   • Download OCC labels from Occ3D-nuScenes

Model Zoo

Method	Dataset	IoU (%)	mIoU (%)	Weights
DRVR	SemanticKITTI	60.3	30.4	Google-drive
DRVR	SemanticPOSS	58.4	22.7	Google-drive
DRVR	nuScenes-Occupancy	30.6	21.3	Google-drive
DRVR	Occ3D (with mask)	84.7	50.1	Google-drive
DRVR	Occ3D (w/o mask)	40.2	29.4	Google-drive

Training and Evaluation

1. SemanticKITTI

   • Modify the configuration file config/drvr_semantickitti.yaml.
   • Enter the tasks/semantickitti_ssc folder, modify run.sh and run the command ./run.sh.

2. SemanticPOSS

   • Modify the configuration file config/drvr_semanticposs.yaml.
   • Enter the tasks/semanticposs_ssc folder, modify run.sh and run the command ./run.sh.

3. nuScenes-Occupancy

   • Modify the configuration file config/drvr_nuscenes.yaml.
   • Enter the tasks/nuscenes_openocc folder, modify run.sh and run the command ./run.sh.

4. Occ3D-nuScenes

   • Modify the configuration file config/drvr_occ3d.yaml.
   • Enter the tasks/nuscenes_occ3d folder, modify run.sh and run the command ./run.sh.