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.

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}}