Efficient Agentic Audio-Visual Question Answering through Toolset Pruning and Modality Consolidation.

arXiv:2610.38960v1 Announce Type: new Abstract: Fine-grained audio-visual question answering (AVQA) requires models to resolve complex cross-modal correspondences and temporally grounding from sparsely distributed evidence. Omni-modal agents, which autonomousl