Home

Awesome

Align-DETR: Improving DETR with Simple IoU-aware BCE loss

Zhi Cai, Songtao Liu, Guodong Wang, Zheng Ge, Xiangyu zhang, Di Huang*

Arxiv version link

Introduction

AlignDETR is a variant of DETR(DEtection with Transformer), with a simple IoU-Aware BCE loss and better performance! It aims to solve the issue of misalignment problem spotted in DETR's output.

<img src="assets/aligndetr.png" >

News 📰

[2024.07.21] We release a stronger version of Align-DETR by making a smooth modification on IA-BCE loss (Now we term as Align Loss in the newest BMVC version)

[2024.07.21] Align-DETR is accepted by BMVC!

Installation 🛠️

Install details can be found in installation instructions

Usage 📖

Train Example

python tools/train_net.py --config-file  aligndetr/aligndetr_k=2_r50_4scale_12ep.py --num-gpus 8

Evaluation Example

python tools/train_net.py --config-file  aligndetr/aligndetr_k=2_r50_4scale_12ep.py --num-gpus 8 --eval train.init_checkpoint=/path/to/checkpoint

Model Zoo 🦁

* represents using a modified IA-BCE loss that absorbs focal loss term.

ModelAPAP50AP75APsAPmAPlweights
AlignDETR-R50-12ep50.367.954.834.153.565.1Google Drive
AlignDETR-R50-24ep51.469.155.835.554.665.7Google Drive
AlignDETR-R50-12ep*50.567.755.334.753.664.6Google Drive
AlignDETR-R50-24ep*51.769.056.335.555.066.1Google Drive

Acknowlegements 🙏

  1. Our code is based on detrex and detectron2.
  2. Align-DETR is also available in the open-source benchmark detrex and mmdetection now!

Citation

If you are interested in our work and use our method in your research, please cite

@misc{cai2023aligndetr,
      title={Align-DETR: Improving DETR with Simple IoU-aware BCE loss}, 
      author={Zhi Cai and Songtao Liu and Guodong Wang and Zheng Ge and Xiangyu Zhang and Di Huang},
      year={2023},
      eprint={2304.07527},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

License

This project is released under the Apache 2.0 license.