• 2025-07-08: URECA training dataset is released!
• 2025-04-08: Our ArXiv Paper is released!
• 🌟 Featured: URECA is now highlighted as a Paper of the Day on Daily Papers page on HuggingFace! 🌟
• 2025-04-06: Training Code, Data collection pipeline, and URECA Model are released.
• 2025-04-06: URECA is released.

Please stay tuned for a URECA Dataset and Evaluation Code!

• Train Code (Apr 6, 2025)
• Pre-trained weights (Apr 6, 2025)
• Code of interactive demo (Apr 6, 2025)
• Demo update (Apr 6, 2025)
• Release ArXiv paper (Apr 8, 2025)
• Training Dataset release (Jul 8, 2025)
• Evaluation Code
• Test Dataset release

conda create -n ureca python=3.9
conda activate ureca

conda install pytorch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 pytorch-cuda=12.4 -c pytorch -c nvidia
pip install -r requirements.txt

Please Download SAM and place it on models folder.
Download URECA Model by following below script.

mkdir models
cd models
git lfs install
git clone https://huggingface.co/SammyLim/URECA
mkdir sam
! Download SAM-H model weight manually!
python gradio_demo/app.py

• Data Curation Pipeline

• URECA Training

We release our URECA training dataset that has 138,152 mask-caption pair! In order to download image-mask pair, please download SA-1B. Get URECA training caption file from Huggingface Link!

Please use the following bibtex to cite our work:

@article{lim2025ureca,
  title={URECA: Unique Region Caption Anything},
  author={Lim, Sangbeom and Kim, Junwan and Yoon, Heeji and Jung, Jaewoo and Kim, Seungryong},
  journal={arXiv preprint arXiv:2504.05305},
  year={2025}
}

This project is largely based on the InternVL repository. Thanks to the authors for their invaluable work and contributions.