Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Skip to content

Voxel Mamba: Group-Free State Space Models for Point Cloud based 3D Object Detection

License

Notifications You must be signed in to change notification settings

gwenzhang/Voxel-Mamba

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Voxel Mamba: Group-Free State Space Models for Point Cloud based 3D Object Detection

arXiv

This repo is the official implementation of the paper Voxel Mamba: Group-Free State Space Models for Point Cloud based 3D Object Detection. Our Voxel Mamba achieves state-of-the-art performance on Waymo and nuScene datasets.

News

-[24-06-18] Voxel Mamba released on arxiv
-[24-07-01] Code of Waymo is released (SOTA).

TODO

  • Release the arXiv version.
  • Clean up and release the code.
  • Release code of Waymo.
  • Release code of NuScenes.
  • Merge Voxel Mamba to OpenPCDet.

Main Results

Waymo Open Dataset

Validation set

Model mAPH_L1 mAPH_L2 Veh_L1 Veh_L2 Ped_L1 Ped_L2 Cyc_L1 Cyc_L2 Log
Voxel Mamba 79.6 73.6 80.8/80.3 72.6/72.2 85.0/80.8 77.7/73.6 78.6/77.6 75.7/74.8 Log

Test set

Model mAPH_L1 mAPH_L2 Veh_L1 Veh_L2 Ped_L1 Ped_L2 Cyc_L1 Cyc_L2 Leaderboard
Voxel Mamba 79.6 74.3 84.4/84.0 77.0/76.6 84.8/80.6 79.0/74.9 75.4/74.3 72.6/71.5 leaderboard

nuScene Dataset

Validation set

Model mAP NDS mATE mASE mAOE mAVE mAAE ckpt Log
Voxel Mamba 67.5 71.9 26.7 25.0 25.8 21.8 18.9 ckpt Log

Test set

Model mAP NDS mATE mASE mAOE mAVE mAAE Leaderboard Submission
Voxel Mamba 69.0 73.0 24.3 23.0 30.9 23.7 13.3 leaderboard Submission

Voxel Mamba's result on Waymo compared with other leading methods. All the experiments are evaluated on an NVIDIA A100 GPU with the same environment. We hope that our Voxel Mamba can provide a potential group-free solution for efficiently handling sparse point clouds for 3D tasks.

Usage

Installation

Please refer to INSTALL.md for installation.

Dataset Preparation

Please follow the instructions from OpenPCDet. We adopt the same data generation process.

Generate Hilbert Template

cd data
mkdir hilbert
python ./tools/hilbert_curves/create_hilbert_curve_template.py

You can also download Hilbert Template files from Google Drive or BaiduYun(code: kubf).

Training

# multi-gpu training
cd tools
bash scripts/dist_train.sh 8 --cfg_file <CONFIG_FILE>

Test

# multi-gpu testing
cd tools
bash scripts/dist_test.sh 8 --cfg_file <CONFIG_FILE> --ckpt <CHECKPOINT_FILE>

Citation

Please consider citing our work as follows if it is helpful.

@article{zhang2024voxel,
  title={Voxel Mamba: Group-Free State Space Models for Point Cloud based 3D Object Detection},
  author={Zhang, Guowen and Fan, Lue and He, Chenhang and Lei, Zhen and Zhang, Zhaoxiang and Zhang, Lei},
  journal={arXiv preprint arXiv:2406.10700},
  year={2024}
}

Acknowledgments

Voxel Mamba is based on OpenPCDet and DSVT.
We also thank the Centerpoint, TransFusion, OctFormer, Mamba, and HEDNet authors for their efforts.

About

Voxel Mamba: Group-Free State Space Models for Point Cloud based 3D Object Detection

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published