Learning Precise Affordances from Egocentric Videos for Robotic Manipulation

Overview

• Code for affordance extraction from egocentric videos is in the folder ego2aff
• Code for affordance model learning is in the folder affordance-learning
• Data: Data_for_Aff-Grasp
• Model and Log: Model_for_Aff-Grasp

Citation

@article{li2025affgrasp,
      title     = {Learning Precise Affordances from Egocentric Videos for Robotic Manipulation}, 
      author    = {Li, Gen and Tsagkas, Nikolaos and Song, Jifei and Mon-Williams, Ruaridh and Vijayakumar, Sethu and Shao, Kun and Sevilla-Lara, Laura},
      journal   = {Proceedings of the IEEE/CVF International Conference on Computer Vision},
      year      = {2025},
    }

Anckowledgement

Part of the code is derived from the hand_object_detector, hoi-forecase, GroundedSAM, and ViT-Adapter. Thanks for their great work!