GSI-Bench

Official Hugging Face organization for GSI-Bench, a benchmark for evaluating generative spatial intelligence in image editing models.

• Paper: arXiv:2604.20570
• Project page: https://aim-uofa.github.io/GSI-Bench/
• Code: https://github.com/aim-uofa/GSI-Bench
• Dataset: https://huggingface.co/datasets/GSI-Bench/GSI-Bench
• BAGEL reproduction artifacts: https://huggingface.co/datasets/GSI-Bench/GSI-Bench-bagel-example
• BAGEL fine-tune checkpoint: https://huggingface.co/GSI-Bench/bagel_finetune_step_10000_nocfg
• Demo: coming soon

Resources

• Evaluation dataset: four official GSI-Bench archives plus a small Hugging Face Dataset Viewer preview split.
• BAGEL artifacts: reproduction bundle for published BAGEL results on the fine split.
• Model checkpoint: BAGEL fine-tune checkpoint for generation experiments.
• Evaluation code: scripts and documentation in the GitHub repository.

Quick Start

Install the Hugging Face CLI:

pip install -U huggingface_hub

Download the official evaluation archives:

hf download GSI-Bench/GSI-Bench \
  fine_dataset.zip mesatask_dataset.zip bathroom_dataset.zip robothor_dataset.zip \
  --repo-type dataset --local-dir GSI-Bench

Download the BAGEL reproduction artifacts:

hf download GSI-Bench/GSI-Bench-bagel-example \
  --repo-type dataset --local-dir bagel_example