ArticleMetadataMain page

source · 2026-04-25

TimeSformer（ICML 2021）：TimeSformer 用纯 attention 结构建模视频时空关系，是视频 Transformer 路线的重要基础。

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

TimeSformer（ICML 2021）：TimeSformer 用纯 attention 结构建模视频时空关系，是视频 Transformer 路线的重要基础。

资料来源：ICML 2021
发表 / 发布日期：2021-07-01
资料加入日期：2026-04-25

一句话结论

TimeSformer 用纯 attention 结构建模视频时空关系，是视频 Transformer 路线的重要基础。

资料定位

这是从体育机器视觉文献清单中拆出的独立来源。当前 first-pass 编译先把它放进 raw/source 证据层，后续可按重要性升级为深分析笔记。

对体育 AI 子线的价值

它为体育视频中的长时序、事件和多主体建模提供通用视频理解背景。

局限或疑问

  • 当前是 bibliography-driven first-pass source note，优先保证来源可追溯和方向定位。
  • 重要 paper 后续应补 analysis.md、关键图页、实验设置和指标细节。

原始材料

  • raw/ingest/2026-04-25-timesformer/source.md
  • raw/ingest/2026-04-25-timesformer/paper-text.md
  • raw/ingest/2026-04-25-timesformer/links.yaml

相关页面

Metadata

{
  "id": "2026-04-25-timesformer",
  "type": "source",
  "title": "TimeSformer（ICML 2021）：TimeSformer 用纯 attention 结构建模视频时空关系，是视频 Transformer 路线的重要基础。",
  "status": "reviewed",
  "created": "2026-04-25",
  "updated": "2026-04-25",
  "venue": "ICML 2021",
  "published_at": "2021-07-01",
  "ingested_at": "2026-04-25",
  "tags": [
    "near-cvpr-2025",
    "video-understanding",
    "reasoning",
    "primary-source"
  ],
  "note_status": "reviewed",
  "source_type": "paper",
  "authors": [],
  "canonical_links": [
    "https://proceedings.mlr.press/v139/bertasius21a/bertasius21a.pdf"
  ],
  "raw_entry": "raw/ingest/2026-04-25-timesformer/",
  "topics": [
    "topics/sports-ai-video-understanding",
    "topics/sports-ai-roadmap",
    "topics/video-understanding"
  ],
  "entities": [
    "entities/sportsmot"
  ],
  "claims": [],
  "questions": []
}