Gen2SM

Generalizable Single-view Object Pose Estimation by Two-side Generating and Matching [WACV 2025 Oral]

ArXiv

( The follow-up work on Extreme-Two-View-Geometry-From-Object-Poses-with-Diffusion-Models )

Setup

Please refer to Extreme-Two-View-Geometry-From-Object-Poses-with-Diffusion-Models

Infer

To eval on the two testsets adopted in E2VG:

python eval_naviTestset.py and python eval_gsoTestset.py

To eval on your custom testset:

1. Refer to Dataset/gso.py or Dataset/navi.py to create a new file implementing CustomDatabase and CustomDataset
2. Run python eval_custom.py. You may modify relevant configurations in eval_custom.py if needed.

Limitation

The current code version assumes that the input images:

• do not exhibit in-plane object rotation
• are captured from viewpoints on the upper hemisphere of the object (i.e., the camera is positioned above the object)

Citation

@InProceedings{sun2024generalizable,
    title     = {Generalizable Single-View Object Pose Estimation by Two-Side Generating and Matching},
    author    = {Sun, Yujing and Sun, Caiyi and Liu, Yuan and Ma, Yuexin and Yiu, Siu Ming},
    booktitle = {Proceedings of the Winter Conference on Applications of Computer Vision (WACV)},
    month     = {February},
    year      = {2025},
    pages     = {545-556}
}