Jinhwan presented “What You Ask is What You Ground: Bridging Question Intent to Temporal Evidence for Grounded VideoQA“
This work addresses question-invariant temporal grounding in grounded VideoQA by conditioning video representations on question intent before localization. It introduces communication-token-based visuo-lingual interaction, factorized MIL cross-attention, and hierarchical multimodal contrastive learning to better align answer prediction with question-specific temporal evidence.