Eff-GRot

Pretrained weights for Eff-GRot: Efficient and Generalizable Rotation Estimation with Transformers (WACV 2026).

• Code and full setup instructions
• Paper

Usage

hf download fanismathioulakis/eff-grot model.safetensors config.json \
  --local-dir ckpts/eff-grot

python eval.py \
  --dataset linemod \
  --checkpoint ckpts/eff-grot/model.safetensors \
  --verbose

The repository defaults reproduce the evaluation settings used in the paper: 64 references, farthest-point reference selection, allocentric poses and in-plane reference augmentation for LINEMOD. See the GitHub repository for CO3D evaluation and dataset preparation.

Citation

@inproceedings{mathioulakis2026eff,
  title={Eff-GRot: Efficient and Generalizable Rotation Estimation with Transformers},
  author={Mathioulakis, Fanis and Radevski, Gorjan and Tuytelaars, Tinne},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  pages={412--421},
  year={2026}
}