Zero-shot 2D Grounding with Novel Affordance Types

2026-08-09

AI Summary

Existing methods struggle to generalize to unseen affordance types, limiting the practical applicability of 2D affordance grounding in real-world scenarios. This work introduces the first zero-shot 2D affordance grounding task, establishes AGD20K-NAT—the first benchmark for this setting—and proposes AffordAnything, a training-free framework that leverages visual segmentation cues and language guidance for inference. Furthermore, the authors present AffordAnything+, a learnable extension that incorporates a trainable cue-fusion module. Experimental results demonstrate that AffordAnything+ outperforms the current state-of-the-art method, OOAL, by an absolute margin of 12.3% on AGD20K-NAT under the IoU@0.4 metric.

Abstract

2D affordance grounding aims to locate the region of an object that a human can interact with. Existing research focuses on recognizing affordance types seen during training and does not study models' ability to generalize to novel affordances, which is crucial for real-world applications. We propose the task of zero-shot 2D grounding with novel affordance types (NAT) and introduce the NAT benchmarks. We then propose AffordAnything, a training-free method that leverages segmentation cues, motivated by the strong correlation between affordance regions and object subparts. To further improve performance, we develop AffordAnything+, a trainable variant that learns to combine these cues. On the proposed AGD20K-NAT benchmark, our best model AffordAnything+ achieves a substantial improvement of 12.3% (absolute) in IoU@0.4 over the SOTA affordance grounding method, OOAL.

Problem

• zero-shot
• 2D grounding
• novel affordance types
• affordance grounding
• generalization

Innovation

• zero-shot affordance grounding
• novel affordance types
• segmentation cues
• AffordAnything
• AGD20K-NAT