Temporal Action Localization在THUMOS14上精度总结

• Efficient Action Detection in Untrimmed Videos via Multi-Task
  Learning 2016 WACV

  THUMOS14 mAP @0.5=19.0%

• CDC: Convolutional-De-Convolutional Networks for Precise Temporal
  Action Localization in Untrimmed Videos 2017 CVPR
  HUMOS14 mAP @0.5=24.7%
  ActvitiyNet1.3mAP=22.9%

• Temporal Action Detection with Structured Segment Networks 2017
  ICCV

  THUMOS14 mAP@0.5=29.8%
  ActvitiyNet1.3 mAP = 28.28%
  代码链接

• Temporal Context Network for Activity Localization in Videos 2017
  ICCV

  THUMOS14 上的mAP@0.5 = 25.6%
  ActvitiyNet 1.3的average mAP为 23.58%

• Cascaded Boundary Regression for Temporal Action Detection2017
  BMVC

  THUMOS14 上的mAP@0.5 为 31.0%· (2017年THUMOS上效果最好效果好,但是也与它用了ActivityNet上预训练的模型有关)
  代码链接

• R-C3D : Region Convolutional 3D Network for Temporal Activity
  Detection 2017 ICCV

  THUMOS14mAP@0.5 =28.9%，
  ActvitiyNet 1.3mAP=26.74%
  代码链接

• Single Shot Temporal Action Detection 2017 ACMMM

  THUMOS14 mAP@0.5 = 24.6%

• BSN: Boundary Sensitive Network for Temporal Action Proposal
  Generation 2018 ECCV

  THUMOS14 mAP @0.5= 36.9%
  代码链接

• End-to-End, Single-Stream Temporal Action Detection in Untrimmed
  Videos 2017 BMVC
  HUMOS14 上的mAP@0.5 = 29.2%
  代码链接

• Exploring Temporal Preservation Networks for Precise Temporal
  Action Localization 2018 AAAI

  THUMOS14 上的mAP@0.5 为 27.6%

• A Self-Adaptive Proposal Model for Temporal Action Detection based
  on Reinforcement Learning 2018 AAAI

  THUMOS14 mAP @0.5=27.7%
  代码链接

• Precise Temporal Action Localization by Evolving Temporal Proposals
  2018 CVPR
  HUMOS14 mAP @0.5= 34.2%

• Rethinking the Faster R-CNN Architecture for Temporal Action
  Localization 2018 CVPR

  THUMOS14 mAP @0.5= 42.0%

• BMN : Boundary-Matching Network for Temporal Action Proposal
  Generation ICCV 2019