SnapSeek 2.0 is a novel system designed for information retrieval from videos participating in the Video Browser Showdown 2025. The system aims to enhance interactivity and accelerate query input speed by employing a human feedback procedure and integrating a magic brush for automatic sketch completion. Building on SnapSeek, which debuted at LSC 2024, the new version retains its predecessor’s most notable features to boost performance including sequencing of scenes in videos and expanding metadata through a variety of computer vision and NLP techniques. To adapt to the competition, specialized datasets including MVK and LapGynLHE undergo detection and segmentation processes, making retrieval more efficient and accurate. Additionally, the system employs a minimalist interface to ensure user-friendliness.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

SnapSeek 2.0 at Video Browser Showdown 2025

  • Minh-Quan Ho-Le,
  • Duy-Khang Ho,
  • Huy-Hoang Do-Huu,
  • Nhut-Thanh Le-Hinh,
  • Hoa-Vien Vo-Hoang,
  • Van-Tu Ninh,
  • Cathal Gurrin,
  • Minh-Triet Tran

摘要

SnapSeek 2.0 is a novel system designed for information retrieval from videos participating in the Video Browser Showdown 2025. The system aims to enhance interactivity and accelerate query input speed by employing a human feedback procedure and integrating a magic brush for automatic sketch completion. Building on SnapSeek, which debuted at LSC 2024, the new version retains its predecessor’s most notable features to boost performance including sequencing of scenes in videos and expanding metadata through a variety of computer vision and NLP techniques. To adapt to the competition, specialized datasets including MVK and LapGynLHE undergo detection and segmentation processes, making retrieval more efficient and accurate. Additionally, the system employs a minimalist interface to ensure user-friendliness.