grt-video-inpaint

GRT Video Inpaint Checkpoint

This repository hosts the released video inpainting checkpoint for Geometric Reciprocity: Unlocking Self-Supervision for Stereoscopic Video Generation (ICML 2026).

• Checkpoint: video_inpaint_large.pth
• Architecture: Video inpainting transformer, large configuration
• Use case: Inpaint right-view disocclusion regions produced by GRT monocular-to-stereo inference
• Code: github.com/Visual-AI/GRT
• Project page: visual-ai.github.io/grt
• Paper: arXiv:2607.05354

Quick Use

git clone https://github.com/Visual-AI/GRT.git
cd GRT
pip install -r requirements.txt
python scripts/inference.py --input-video assets/example.mp4 --output-dir output/example --output-mode all

The inference script downloads this checkpoint automatically on first run.

Citation

@inproceedings{lu2026grt,
  author    = {Jingyi Lu and Kai Han},
  title     = {Geometric Reciprocity: Unlocking Self-Supervision for Stereoscopic Video Generation},
  booktitle = {International Conference on Machine Learning (ICML)},
  year      = {2026},
}