ArticleMetadataMain page

source · 2026-04-25

MultiSports（ICCV 2021）：多人时空动作定位的数据锚点

#near-cvpr-2025#video-understanding#benchmarking#primary-source

MultiSports（ICCV 2021）：多人时空动作定位的数据锚点

会议 / 来源：ICCV 2021
发表日期：2021-10-01
资料加入日期：2026-04-25

一句话结论

面向多人、多运动、时空动作定位，提供 clips、动作实例和边框标注，补上体育 AI 的动作定位基准层。

论文定位

MultiSports 将体育动作理解推进到多人、多运动、tube-level / box-level 的时空定位问题，使模型必须同时处理动作类别、主体位置和时间边界。

对体育 AI 子线的价值

它补强了当前 wiki 在 tracking 之外的 action localization 证据，避免体育 AI 子线过度集中在 MOT。

和现有证据的关系

局限或疑问

  • 当前这版 source note 先完成 bibliography-driven 编译，后续可以按 deep-analysis-first 流程补 analysis.md、关键图页和实验细节。
  • 具体数值和 benchmark 细节应以 raw/ingest/2026-04-25-multisports/paper-text.md 与原始 PDF 为准。

原始材料

  • raw/ingest/2026-04-25-multisports/paper.pdf
  • raw/ingest/2026-04-25-multisports/paper-text.md
  • raw/ingest/2026-04-25-multisports/abstract.md
  • raw/ingest/2026-04-25-multisports/links.yaml

相关页面

Metadata

{
  "id": "2026-04-25-multisports",
  "type": "source",
  "title": "MultiSports（ICCV 2021）：多人时空动作定位的数据锚点",
  "status": "reviewed",
  "created": "2026-04-25",
  "updated": "2026-04-25",
  "venue": "ICCV 2021",
  "published_at": "2021-10-01",
  "ingested_at": "2026-04-25",
  "tags": [
    "near-cvpr-2025",
    "video-understanding",
    "benchmarking",
    "primary-source"
  ],
  "note_status": "reviewed",
  "source_type": "paper",
  "authors": [],
  "canonical_links": [
    "https://openaccess.thecvf.com/content/ICCV2021/html/Li_MultiSports_A_Multi-Person_Video_Dataset_of_SpatioTemporally_Localized_Sports_Actions_ICCV_2021_paper.html",
    "https://openaccess.thecvf.com/content/ICCV2021/papers/Li_MultiSports_A_Multi-Person_Video_Dataset_of_Spatio-Temporally_Localized_Sports_Actions_ICCV_2021_paper.pdf"
  ],
  "raw_entry": "raw/ingest/2026-04-25-multisports/",
  "topics": [
    "topics/sports-ai-video-understanding",
    "topics/sports-ai-roadmap",
    "topics/video-understanding"
  ],
  "entities": [
    "entities/sportsmot"
  ],
  "claims": [],
  "questions": []
}