作者：Rheelt
项目：Materials-Temporal-Action-Detection

本文汇总了时序行为检测相关的资源列表

更多Awsome Github资源请关注：【Awsome】GitHub 资源汇总

Papers: temporal action proposals & detection

• (TGM) Temporal Gaussian Mixture Layer for Videos (ICML 2019) CODE.pytorch
• (MGG) Multi-granularity Generator for Temporal Action Proposal (CVPR 2019)
• (GTAN) Gaussian Temporal Awareness Networks for Action Localization (CVPR 2019)

Papers: weakly temporal action detection

• (CMCS) Completeness Modeling and Context Separation for Weakly Supervised Temporal Action Localization (CVPR19)CODE.pytorch
• Weakly-Supervised Temporal Localization via Occurrence Count Learning (ICML 2019)
• (MAAN) Marginalized Average Attentional Network for Weakly-Supervised Learning (ICLR2019)CODE.pytorch
• (WSGN) Weakly Supervised Gaussian Networks for Action Detection (Arxiv 2019.4)
• (RefineLoc) RefineLoc: Iterative Refinement for Weakly-Supervised Action Localization (Arxiv 2019.4)
• (STAR) Segregated Temporal Assembly Recurrent Networks for Weakly Supervised Multiple Action Detection (AAAI 2019)
• (TSRNet) Learning Transferable Self-attentive Representations for Action Recognition in Untrimmed Videos with Weak Supervision (AAAI 2019)
• (StepByStep) Step-by-step Erasion, One-by-one Collection: AWeakly Supervised Temporal Action Detector (MM 2018)
• (W-TALC) W-TALC: Weakly-supervised Temporal Activity Localization and Classification (ECCV 2018) CODE.pytorch
• (AutoLoc) AutoLoc: Weakly-supervised Temporal Action Localization in Untrimmed Videos (ECCV 2018) CODE.caffe
• (STPN) Weakly Supervised Action Localization by Sparse Temporal Pooling Network (CVPR 2018) CODE.tensorflow.unofficial
• (H&S) Hide-and-seek: Forcing a network to be meticulous for weakly-supervised object and action localization (ICCV 2017)
• (UNet) UntrimmedNets for Weakly Supervised Action Recognition and Detection (CVPR 2017) CODE.caffe

Features: Download link

THUMOS14

• C3D: link
• I3D: Video is sampled at 25 frames per second. 16 frames as a video unit. link
• UNet: link
• ANet2016-cuhk(4096dims): 6 frames as a video unit. link
• ANet2016-cuhk(3072dims): 5 frames as a video unit. link

ActivityNet v1.2

• I3D: Video is sampled at 25 frames per second. 16 frames as a video unit. link
• UNet: link

ActivityNet v1.3

• C3D: link
• ANet2016-cuhk(400dims): 16 frames as a video unit. link
• I3D: 16 frames as a video unit. link
• ANet2016-cuhk(3072dims): 16 frames as a video unit. link

Benchmark Results (THUMOS14 Results)

These methods are listed in chronological order.

Method Feature IoU-> 0.1 0.2 0.3 0.4 0.5 0.6 0.7 MAAN I3D

59.8 50.8 41.1 30.6 20.3 12.0 6.9 CMCS I3D

57.4 50.8 41.2 32.1 23.1 15.0 7.0 WSGN I3D

55.3 47.6 38.9 30.0 21.1 13.9 8.3 RefineLoc UNet

33.9

22.1

6.1 STAR I3D

68.8 60.0 48.7 34.7 23.0

TSRNet 2-Stream(ResNet101)

55.9 46.9 38.3 28.1 18.6 11.0 5.59 StepByStep TSN

45.8 39.0 31.1 22.5 15.9

W-TALC UNet

49.0 42.8 32.0 26.0 18.8

6.2 W-TALC I3D

55.2 49.6 40.1 31.1 22.8

7.6 AutoLoc UNet

35.8 29.0 21.2 13.4 5.8 STPN I3D

52.0 44.7 35.5 25.8 16.9 9.9 4.3 H&S C3D

36.4 27.8 19.5 12.7 6.8

UNet UNet

44.4 37.7 28.2 21.1 13.7

推荐阅读：
【资源】常用的语义分割架构结构综述以及代码复现
【资源】视频分析 / 多模态学习论文、代码、数据集大列表

微信公众号: 极市平台（ID: extrememart ）
每天推送最新CV干货