Affordance Transfer Learning for Human-Object Interaction Detection
This work addresses the challenge of scene understanding in computer vision by enabling more robust HOI detection with novel objects, which is incremental but impactful for applications like robotics and human-computer interaction.
The paper tackles the problem of detecting human-object interactions (HOI) with novel objects by introducing an affordance transfer learning approach that decouples HOI representations into affordance and object components, enabling composition of novel interactions and improving HOI detection performance, especially for unseen objects, with significant gains over state-of-the-art methods on HICO-DET and HOI-COCO datasets.
Reasoning the human-object interactions (HOI) is essential for deeper scene understanding, while object affordances (or functionalities) are of great importance for human to discover unseen HOIs with novel objects. Inspired by this, we introduce an affordance transfer learning approach to jointly detect HOIs with novel objects and recognize affordances. Specifically, HOI representations can be decoupled into a combination of affordance and object representations, making it possible to compose novel interactions by combining affordance representations and novel object representations from additional images, i.e. transferring the affordance to novel objects. With the proposed affordance transfer learning, the model is also capable of inferring the affordances of novel objects from known affordance representations. The proposed method can thus be used to 1) improve the performance of HOI detection, especially for the HOIs with unseen objects; and 2) infer the affordances of novel objects. Experimental results on two datasets, HICO-DET and HOI-COCO (from V-COCO), demonstrate significant improvements over recent state-of-the-art methods for HOI detection and object affordance detection. Code is available at https://github.com/zhihou7/HOI-CL