We have now released the full sets (trainval, test) of AI-TOD-v2! [Download]
This is a repository of the official implementation of the following paper:
- [Paper][Code] Detecting tiny Objects in aerial images: A normalized Wasserstein distance and A new benchmark (ISPRS J P & RS, 2022)
- [Paper][Code] Dot distance for tiny object detection in aerial images (CVPRW, 2021)
The Normalized Wasserstein Distance and the RanKing-based Assigning strategy (NWD-RKA) for tiny object detection.
A comparison between AI-TOD and AI-TOD-v2.
Notes: The images of the AI-TOD-v2 are the same of the AI-TOD. In this stage, we only release the train, val annotations of the AI-TOD-v2, the test annotations will be used to hold further competitions.
Supported baselines for tiny object detection:
Supported horizontal tiny object detection methods:
Supported rotated tiny object detection methods:
Required environments:
- Linux
- Python 3.6+
- PyTorch 1.3+
- CUDA 9.2+
- GCC 5+
- MMCV
- cocoapi-aitod
Install TODbox:
Note that our TODbox is based on the MMDetection 2.24.1. Assume that your environment has satisfied the above requirements, please follow the following steps for installation.
git clone https://github.com/Chasel-Tsui/mmdet-aitod.git
cd mmdet-nwdrka
pip install -r requirements/build.txt
python setup.py develop
If you use this repo in your research, please consider citing these papers.
@inproceedings{xu2021dot,
title={Dot Distance for Tiny Object Detection in Aerial Images},
author={Xu, Chang and Wang, Jinwang and Yang, Wen and Yu, Lei},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
pages={1192--1201},
year={2021}
}
@inproceedings{NWDRKA_2022_ISPRS,
title={Detecting Tiny Objects in Aerial Images: A Normalized Wasserstein Distance and A New Benchmark},
author={Xu, Chang and Wang, Jinwang and Yang, Wen and Yu, Huai and Yu, Lei and Xia, Gui-Song},
booktitle={ISPRS Journal of Photogrammetry and Remote Sensing},
volume={190},
pages={79--93},
year={2022},
}