RSDet-datasets

# RSDet-datasets RSDet-datasets is a curated collection of 33 widely used remote sensing object detection datasets, totaling 700+ GB. It provides centralized access to these datasets with unified DOTA-style TXT and COCO JSON annotations, supporting both : - horizontal bounding box (HBB) - oriented bounding box (OBB) ## Download Download a specific dataset using the Hugging Face CLI: ``` hf download VisionXLab/RSDet-datasets \ --include "your_dataset/**" \ --local-dir /your/local/dir \ --repo-type dataset ``` Replace your_dataset with the name of the dataset you want to download. ## Dataset Structure Each dataset directory includes a `README.md`, either in the dataset root or its `process/` directory, describing the directory structure and preprocessing details when applicable. Preprocessing code is also provided to convert the original annotations into the unified: * DOTA-style TXT * COCO JSON Small processed examples for both annotation formats are provided under processed_meta/ for quick inspection. <p align="center"> <img src="DIOR/processed_meta/vis_hbb_obb/16604.jpg" width="600"> </p> ## Citation If you find RSDet-datasets useful in your research, please consider citing this repository: ``` @misc{visionxlab2026rsdetdatasets, title = {RSDet-datasets: A Collection of Remote Sensing Object Detection Datasets}, author = {Shuran Ma and Qingyun Li and Xue Yang}, year = {2026}, publisher = {VisionXLab}, url = {https://huggingface.co/datasets/VisionXLab/RSDet-datasets}} ```