https://github.com/LiewFeng/RayDN
[ECCV 2024] Ray Denoising (RayDN): Depth-aware Hard Negative Sampling for Multi-view 3D Object Detection
https://github.com/LiewFeng/RayDN
3d-object-detection autonomous-driving denoising eccv eccv2024 nuscenes
Last synced: 7 months ago
JSON representation
[ECCV 2024] Ray Denoising (RayDN): Depth-aware Hard Negative Sampling for Multi-view 3D Object Detection
- Host: GitHub
- URL: https://github.com/LiewFeng/RayDN
- Owner: LiewFeng
- License: other
- Created: 2024-02-03T08:59:17.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2024-09-30T03:33:23.000Z (about 1 year ago)
- Last Synced: 2024-10-30T04:11:31.822Z (12 months ago)
- Topics: 3d-object-detection, autonomous-driving, denoising, eccv, eccv2024, nuscenes
- Language: Python
- Homepage: https://arxiv.org/abs/2402.03634
- Size: 21 MB
- Stars: 72
- Watchers: 2
- Forks: 1
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
RayDN
Ray Denoising: Depth-aware Hard Negative Sampling for Multi-view 3D Object Detection
[](https://paperswithcode.com/sota/3d-object-detection-on-nuscenes-camera-only?p=beam-beta-distribution-ray-denoising-for)
[](https://arxiv.org/abs/2402.03634)
[](https://github.com/LiewFeng/RayDN/assets/42316773/de2c229b-0f6a-4456-b72c-508ea6161198)
## Introduction
This repository is an official implementation of our ***ECCV 2024*** paper [Ray Denoising: Depth-aware Hard Negative Sampling for Multi-view 3D Object Detection](https://arxiv.org/abs/2402.03634). This repository contains Pytorch training code, evaluation code and pre-trained models.
## Framework
## Getting Started
Our code is built based on [StreamPETR](https://github.com/exiawsh/StreamPETR). Please follow [StreamPETR](https://github.com/exiawsh/StreamPETR) to [setup enviroment](https://github.com/exiawsh/StreamPETR/blob/main/docs/setup.md) and [prepare data](https://github.com/exiawsh/StreamPETR/blob/main/docs/data_preparation.md) step by step.
## Training and Inference
You can train the model following:
```angular2html
tools/dist_train.sh projects/configs/RayDN/raydn_eva02_800_bs2_seq_24e.py 8
```
You can evaluate the detection model following:
```angular2html
tools/dist_test.sh projects/configs/RayDN/raydn_eva02_800_bs2_seq_24e.py work_dirs/raydn_eva02_800_bs2_seq_24e/latest.pth 8 --eval bbox
```
## Results on NuScenes Val Set.
| Model | Setting |Pretrain| Lr Schd | NDS| mAP| Config | Download |
| :---: | :---: | :---: | :---: | :---:|:---:| :---: | :---: |
| RayDN | R50 - 428q | NuImg | 60ep | 56.1 | 47.1 | [config](projects/configs/RayDN/raydn_r50_flash_704_bs2_seq_428q_nui_60e.py) | [ckpt](https://mailsucasaccn-my.sharepoint.com/:u:/g/personal/liufeng20_mails_ucas_ac_cn/EYtElqwLxxRMqewZ0qZIz2wBmfLoPrOe3YIVdlLVZSKGcQ?e=wdbkHi) |
| RayDN | EVA02-L - 900q | EVA02 | 24ep | 62.4 | 54.1 | [config](projects/configs/RayDN/raydn_eva02_800_bs2_seq_24e.py) |[ckpt](https://mailsucasaccn-my.sharepoint.com/:u:/g/personal/liufeng20_mails_ucas_ac_cn/ERYKTAGGSKRFmrDoF6VnUf8BKw96Cw-rNyvbFFrouQWJBw?e=Dkcil3) |
## Acknowledgements
We thank these great works and open-source codebases:
[MMDetection3d](https://github.com/open-mmlab/mmdetection3d), [StreamPETR](https://github.com/exiawsh/StreamPETR), [DETR3D](https://github.com/WangYueFt/detr3d), [PETR](https://github.com/megvii-research/PETR).
## Citation
If you find RayDN is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry.
```bibtex
@article{liu2024ray,
title={Ray Denoising: Depth-aware Hard Negative Sampling for Multi-view 3D Object Detection},
author={Liu, Feng and Huang, Tengteng and Zhang, Qianjing and Yao, Haotian and Zhang, Chi and Wan, Fang and Ye, Qixiang and Zhou, Yanzhao},
journal={arXiv preprint arXiv:2402.03634},
year={2024}
}
```