AVSD

The AVSD (Audio-Visual Scene Description) dataset is a multimodal benchmark that contains audio-visual data used to evaluate the reasoning and response generation capabilities of models in understanding and integrating sensory information from complex environments.

Papers using AVSD (3)

Not All Modalities Are Equal: Instruction-Aware Gating for Multimodal Videos2026Chain of Questions: Guiding Multimodal Curiosity in Language Models2025End-to-End Multimodal Representation Learning for Video Dialog2022 · 1 cites