HLR-AVSceneQA
A hierarchical evaluation benchmark for spatial audio-visual scene understanding, designed around three progressively ordered capability levels: Hear, Localize and Reason. It distinguishes between egocentric and allocentric/environment-centric frames of reference, allowing diagnosis of bottlenecks at each level of a model.
Timeline (last 90 days)
The research team proposed HLR-AVSceneQA, a comprehensive spatial audio-visual evaluation benchmark designed according to the 'Hear-Localize-Reason' hierarchical structure.
The three-layer structure of HLR-AVSceneQA corresponds to three progressive capabilities: hearing (recognizing sound content), localizing (localizing sound sources under egocentric and allocentric perspectives), and reasoning (reasoning about the relationships between sound sources and other elements in the visual scene).
All Facts (2)
The three-layer structure of HLR-AVSceneQA corresponds to three progressive capabilities: hearing (recognizing sound content), localizing (localizing sound sources under egocentric and allocentric perspectives), and reasoning (reasoning about the relationships between sound sources and other elements in the visual scene).
50%UnverifiedThe research team proposed HLR-AVSceneQA, a comprehensive spatial audio-visual evaluation benchmark designed according to the 'Hear-Localize-Reason' hierarchical structure.
50%