Home

Awesome

DQ-DETR: DETR with Dynamic Query for Tiny Object Detection

method

News

[2024/12/06] We released the organized datasets AI-TOD-V1 and AI-TOD-V2.

[2024/7/1]: DQ-DETR has been accepted by ECCV 2024. πŸ”₯πŸ”₯πŸ”₯

[2024/5/3]: DNTR has been accepted by TGRS 2024. πŸ”₯πŸ”₯πŸ”₯

Installation -- Compiling CUDA operators

conda create -n dqdetr python=3.9 --y
conda activate dqdetr
bash install.sh
<!-- # bash scripts/DQ_eval.sh /nfs/home/hoiliu/dqdetr/weights/dqdetr_best305.pth -->

Eval models

bash scripts/DQ_eval.sh /path/to/your/dataset /path/to/your/checkpoint

Trained Model

CUDA_VISIBLE_DEVICES=5,6,7 bash scripts/DQ.sh /path/to/your/dataset

Our works on Tiny Object Detection

TitleVenueLinks
DNTRTGRS 2024Paper | code
DQ-DETRECCV 2024Paper | code

Performance

Table 1. Training Set: AI-TOD-V2 trainval set, Testing Set: AI-TOD-V2 test set, 36 epochs, where FRCN, DR denotes Faster R-CNN and DetectoRS, respectively.

MethodBackbonemAPAP<sub>50</sub>AP<sub>75</sub>AP<sub>vt</sub>AP<sub>t</sub>AP<sub>s</sub>AP<sub>m</sub>
Faster R-CNNR-5011.126.37.60.07.223.333.6
NWD-RKAR-5023.453.516.88.723.828.536.0
DAB-DETRR-5022.455.614.39.021.728.338.7
DINO-DETRR-5025.961.317.512.725.332.039.7
DQ-DETRR-5030.569.222.715.230.936.845.5

AI-TOD-v1 and AI-TOD-v2 Datasets

https://drive.google.com/drive/folders/1hkbcZ3TPABx3QxoCufE1KAPu55Ibw-8d?usp=sharing
β”œβ”€ Dataset
β”‚   └─ aitod
β”‚       β”œβ”€ annotations
β”‚       β”œβ”€ images
β”‚       β”œβ”€ test
β”‚       β”œβ”€ train
β”‚       β”œβ”€ trainval
β”‚       └─ val
β”œβ”€ DQ-DETR

Pretrained Weights

Citation


@InProceedings{huang2024dq,
author={Huang, Yi-Xin and Liu, Hou-I and Shuai, Hong-Han and Cheng, Wen-Huang},
title={DQ-DETR: DETR withΒ Dynamic Query forΒ Tiny Object Detection},
booktitle={European Conference on Computer Vision},
pages={290--305},
year={2025},
organization={Springer}
}

@ARTICLE{10518058,
  author={Liu, Hou-I and Tseng, Yu-Wen and Chang, Kai-Cheng and Wang, Pin-Jyun and Shuai, Hong-Han and Cheng, Wen-Huang},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={A DeNoising FPN With Transformer R-CNN for Tiny Object Detection}, 
  year={2024},
  volume={62},
  number={},
  pages={1-15},
}