projects/dab_detr/README.md

DAB-DETR: Dynamic Anchor Boxes are Better Queries for DETR

Shilong Liu, Feng Li, Hao Zhang, Xiao Yang, Xianbiao Qi, Hang Su, Jun Zhu, Lei Zhang

[arXiv] [BibTeX]

Pretrained Models

Here we provide the pretrained DAB-DETR weights based on detrex.

Name	Backbone	Pretrain	Epochs	box AP	download
DAB-DETR-R50	R-50	IN1k	50	43.3	model
DAB-DETR-R101	R-101	IN1k	50	44.0	model
DAB-DETR-Swin-T	Swin-T	IN1k	50	45.2	model

Converted Models

Here are the converted the pretrained weights from DAB-DETR official repo.

Name	Backbone	Pretrain	Epochs	box AP	download
DAB-DETR-R50-3patterns	R-50	IN1k	50	42.8	model
DAB-DETR-R50-DC5	R-50	IN1k	50	44.6	model
DAB-DETR-R50-DC5-3patterns	R-50	IN1k	50	45.7	model
DAB-DETR-R101-DC5	R-101	IN1k	50	45.7	model

Training

All configs can be trained with:

cd detrex
python tools/train_net.py --config-file projects/dab_detr/configs/path/to/config.py --num-gpus 8

By default, we use 8 GPUs with total batch size as 16 for training.

Evaluation

Model evaluation can be done as follows:

cd detrex
python tools/train_net.py --config-file projects/dab_detr/configs/path/to/config.py --eval-only train.init_checkpoint=/path/to/model_checkpoint

Citing DAB-DETR

If you find our work helpful for your research, please consider citing the following BibTeX entry.

@inproceedings{
      liu2022dabdetr,
      title={{DAB}-{DETR}: Dynamic Anchor Boxes are Better Queries for {DETR}},
      author={Shilong Liu and Feng Li and Hao Zhang and Xiao Yang and Xianbiao Qi and Hang Su and Jun Zhu and Lei Zhang},
      booktitle={International Conference on Learning Representations},
      year={2022},
      url={https://openreview.net/forum?id=oMI9PjOb9Jl}
}