Summarize Before Grounding: Query-Guided Chunk Condensation for Long-Video Temporal Grounding
This paper proposes a ``summarize before grounding''framework (named ``SumGround'') for long-video temporal grounding, and introduces query-guided latent summaries, which is represented as KV states of query-guided prompts, to compress redundant visual tokens into compact query-relevant chunk summaries.
Nan-Xing Hu, Xiao-Yue Duan, Qi-Wei Yan et al.
· 0 citations