Let Occ Flow: Self-Supervised 3D Occupancy Flow Prediction

Paper | Project Page

Let Occ Flow: Self-Supervised 3D Occupancy Flow Prediction, CoRL 2024

Yili Liu*, Linzhan Mou*, Xuan Yu, Chenrui Han, Sitong Mao, Rong Xiong, Yue Wang$\dagger$

* Equal contribution $\dagger$ Corresponding author

Update

• [2025/04/16] - Code released.
• [2024/09/04] - Our paper has been accepted to CoRL 2024. We will release the code in this repository.
• [2024/07/18] - We released our paper on arXiv.

Demo

Results on KITTI-MOT dataset:

Results on nuScenes dataset:

Introduction

• We proposed Let Occ Flow, the first self-supervised method for jointly predicting 3D occupancy and occupancy flow, by integrating 2D optical flow cues into geometry and motion optimization.
• We designed a novel attention-based temporal fusion module for efficient temporal interaction. Furthermore, we proposed a flow-oriented optimization strategy to mitigate the training instability and sample imbalance problem.
• We conducted extensive experiments on various datasets with qualitative and quantitative analyses to show the competitive performance of our approach.

Getting Started

Installation

Follow detailed instructions in Installation.

Preparing Dataset

Follow detailed instructions in Prepare Dataset.

Train

# kitti odometry(tab.1)
# train occ model
python train.py --py-config config/kitti/kitti_occ_odom.py --work-dir out/train/kitti/occ_odom --dataset kitti --depth-metric 

# nuscenes(tab.3)
# train occ model
python train.py --py-config config/nuscenes/nuscenes_occ_voxelaffm.py --work-dir out/train/nuscenes/occ_static_train --dataset nuscenes --depth-metric

# train occ flow model (modify the parameter 'load_from' in the config file)
python train.py --py-config config/nuscenes/nuscenes_occ_flow_voxelaffm.py --work-dir out/train/nuscenes/occ_flow_train --dataset nuscenes --depth-metric 

Evaluation and Visualization

# save occupancy, occupancy flow and render results
# (modify the parameter 'load_from' in the config file)
python eval.py --py-config config/kitti/kitti_occ_odom.py --work-dir out/visualization/kitti/kitti_odom --resolution 0.2 --dataset kitti

python eval.py --py-config config/nuscenes/nuscenes_occ_flow_voxelaffm.py --work-dir out/visualization/nuscenes/occ_flow --resolution 0.4 --dataset nuscenes

# prepare ray casting for ray_iou_geo metric
python utils/ray_iou_geo/ray_casting_kitti.py --pred-occ-path out/visualization/nuscenes/kitti/kitti_odom --output-dir /path/to/project/ray_iou_output/kitti_odom

python utils/ray_iou_geo/ray_casting_nus.py --pred-occ-path out/visualization/nuscenes/occ_flow/occupancy --output-dir /path/to/project/ray_iou_output/occ_flow

# eval ray_iou_geo metric
python utils/ray_iou_geo/metric_kitti.py --work-dir /path/to/project/ray_iou_output/kitti_odom

python utils/ray_iou_geo/metric.py --work-dir /path/to/project/ray_iou_output/occ_flow

# visualize occupancy and occupancy flow
python visualize_occupancy.py

Acknowledgement

Many thanks to these excellent projects.

• SelfOcc
• OccNeRF
• OccNet
• Unimatch
• Cotracker

Citation

If this work is helpful for your research, please consider citing the following paper:

@article{liu2024letoccflow,
    title={Let Occ Flow: Self-Supervised 3D Occupancy Flow Prediction},
    author={Yili Liu and Linzhan Mou and Xuan Yu and Chenrui Han and Sitong Mao and Rong Xiong and Yue Wang},
    journal={arXiv preprint arXiv:2407.07587},
    year={2024},
}