Knowledge graphs for physical affordance-aware object detection

Integrate knowledge graphs encoding physical affordances, such as typical support relations between objects, into object-detection pipelines.

Background

The survey discusses the limitations of conventional object detectors that predict object categories and bounding boxes without modeling physical or spatial relations. Knowledge about affordances and physically plausible configurations could help detectors distinguish coherent scenes from situations involving unsupported or implausibly positioned objects. The paper explicitly identifies incorporating such knowledge into detection systems as an unresolved research direction.

References

Integrating knowledge graphs of physical affordances (e.g., cups usually sit on flat surfaces) into detection pipelines remains an open research direction.

— Commonsense Reasoning in Computer Vision: Foundations, Recent Advancements, and Future Directions  (2609.05257 - Mahmud et al., 4 Sep 2026) in Section 4.2, Object Detection: Spatial Relations and Affordances