Do All Visual Tokens Matter Equally? Object-Evidence Preserving Token Merging for Vision-Language Retrieval

Do All Visual Tokens Matter Equally? Object-Evidence Preserving Token Merging for Vision-Language Retrieval

Object-aware token merging framework SaMer compresses image-side tokens while preserving query-selectable visual evidence, achieving significant storage reduction and improved retr…

Hugging Face · Daily Papers ·Suhyeong Park, Junha Jung · ·▲ 11 upvotes

Este artigo está em destaque na seleção diária de papers do Hugging Face, curada pela comunidade de pesquisa em IA.

Autores: Suhyeong Park, Junha Jung, Jungwoo Park, Jaewoo Kang

  • 11 upvotes da comunidade
  • Temas: vision-language retrieval, late interaction, token compression, object-aware merging, post-projector tokens, centroid compression

Resumo

Resumo original (em inglês), extraído do paper:

Object-aware token merging framework SaMer compresses image-side tokens while preserving query-selectable visual evidence, achieving significant storage reduction and improved retrieval performance.

Onde ler

compartilhar: