https://github.com/zehuichen123/autoalignv2

[ECCV2022, IJCAI2022] AutoAlignV2: Deformable Feature Aggregation for Dynamic Multi-Modal 3D Object Detection
https://github.com/zehuichen123/autoalignv2

Last synced: about 2 months ago
JSON representation

[ECCV2022, IJCAI2022] AutoAlignV2: Deformable Feature Aggregation for Dynamic Multi-Modal 3D Object Detection

• Host: GitHub
• URL: https://github.com/zehuichen123/autoalignv2
• Owner: zehuichen123
• License: apache-2.0
• Created: 2022-06-02T01:39:05.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2022-10-14T05:48:26.000Z (about 2 years ago)
• Last Synced: 2024-10-28T06:58:06.830Z (2 months ago)
• Language: Python
• Homepage:
• Size: 15.7 MB
• Stars: 144
• Watchers: 11
• Forks: 15
• Open Issues: 8
• Metadata Files:
  • Readme: readme.md
  • License: LICENSE

Awesome Lists containing this project

README

        


 AutoAlignV2 

Deformable Feature Aggregation for Dynamic Multi-Modal 3D Object Detection


Zehui Chen, Zhenyu Li, Shiquan Zhang, Liangji Fang, Qinhong Jiang, Feng Zhao. 




[ECCV 2022 Paper] 
 






## Performance

### nuScenes Val set

| Model | config | mAP | NDS |

| - | - | - | - |

| Baseline 1/8 subset | | - | - |

| AutoAlignV2 1/8 subset | | 58.5 | 63.2 |

### nuScenes Test Leaderboard

| Model | mAP | NDS |

| -|-|-|

| AutoAlign |  65.8  |  70.9  |

| AutoAlignV2 | 68.4 | 72.4|

## Note

The code is released without rechecking. We will clean up and recheck the code recently.

## Get Started

### Install Deformable Ops from DeformDETR

```

cd ops

sh ./make.sh

# unit test (should see all checking is True)

python test.py

```

### Prepare Dataset

```

python tools/create_data.py nuscenes --root-path ./data/nuscenes --out-dir ./data/nuscenes --extra-tag nuscenes --version v1.0

```

### Train Model

```

./tools/dist_train.sh aav2_cfg/centerpoint_voxel_nus_8subset_bs4_img1_nuimg_detach_deform_multipts.py 8

```

### Test Model

```

./tools/dist_test.sh aav2_cfg/centerpoint_voxel_nus_8subset_bs4_img1_nuimg_detach_deform_multipts.py work_dirs/centerpoint_voxel_nus_8subset_bs4_img1_nuimg_detach_deform_multipts/epoch_20.pth 8 --eval bbox

```

### Pretrained Model Weights

| Model | Weights |

| - | -|

|Pretrained YOLOX-640 | https://drive.google.com/file/d/1DuA16ocF8Dmfl0nC3NNSKG_Yqq-aXuLH/view?usp=sharing |

|Pretrained YOLOX-800 | https://drive.google.com/file/d/1YPSjG38qXrV8gYlq5FuhQCjfRIUvhaju/view?usp=sharing |

| AutoAlignV2 1/8 subset| https://drive.google.com/file/d/1F_B4bq5Y6QdkUHQbqeLaEGl-IZHHe1WB/view?usp=sharing |

## Citation

If you find our work useful for your research, please consider citing the paper

```

@article{chen2022autoalignv2,

  title={AutoAlignV2: Deformable Feature Aggregation for Dynamic Multi-Modal 3D Object Detection},

  author={Chen, Zehui and Li, Zhenyu and Zhang, Shiquan and Fang, Liangji and Jiang, Qinhong and Zhao, Feng},

  journal={ECCV},

  year={2022}

}

```