视频行为识别与轻量化网络的前沿论文、代码等
https://zhuanlan.zhihu.com/c_1207774575393865728
CVPR 2020 行为识别/视频理解论文汇总
https://zhuanlan.zhihu.com/p/141429177
CVPR 2020 论文大盘点-动作识别篇
https://cloud.tencent.com/developer/article/1664055
CVPR 2020 论文大盘点-动作检测与动作分割篇
https://www.sohu.com/a/408454247_823210
Shift思想在视频理解中的近期进展
https://zhuanlan.zhihu.com/p/137385332
视频行为识别2020最新研究进展(中科院深圳先进技术研究院-乔宇)
https://zhuanlan.zhihu.com/p/109519047
ECCV 2020 论文大盘点-动作检测与识别篇
https://blog.csdn.net/moxibingdao/article/details/109140629
Temporal Action Detection总结
https://zhuanlan.zhihu.com/p/52524590
https://www.zhihu.com/question/57523080/answer/158568414
https://zhuanlan.zhihu.com/p/26603387
https://blog.csdn.net/qq_33278461/article/details/80720104
时序行为检测 & 弱监督时序行为检测 & 时序行为proposal生成 论文整理
https://zhuanlan.zhihu.com/p/112811396?utm_source=wechat_session
CVPR2019 | 论文之行为/动作识别、手势识别、时序动作检测及视频相关
https://blog.csdn.net/leiduifan6944/article/details/109624879
ECCV CVPR AAAI 2018年 Action recognition 的汇总
https://www.sohu.com/a/298599618_100021558
2018年 Action recognition 的汇总(ECCV CVPR AAAI)
https://zhuanlan.zhihu.com/p/56061717
Action Localization Benchmarks
Papers and Results of Temporal Action Localization
https://github.com/VividLe/awesome-weakly-supervised-action-localization
Papers: temporal action proposals & detection
Papers: weakly temporal action detection
Features: Download link
Benchmark Results (THUMOS14 Results)
https://github.com/sming256/Materials-Temporal-Action-Detection
AVA数据集:
https://zhuanlan.zhihu.com/p/157869607
时空行为定位相关论文:
https://blog.csdn.net/irving512?t=1
人类动作识别数据集AVA:
https://blog.csdn.net/zchang81/article/details/78291527?utm_medium=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-3.channel_param&depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromMachineLearnPai2-3.channel_param
https://blog.csdn.net/gh13uy2ql0n5/article/details/78302372?utm_medium=distribute.pc_relevant.none-task-blog-title-2&spm=1001.2101.3001.4242
下载地址:
https://research.google.com/ava/
视频特征提取工具:
(I3D models trained on Kinetics)
https://github.com/piergiaj/pytorch-i3d
一、行为识别:
(1)ECCV2020 腾讯优图 时间差异表示学习
Temporal Distinct Representation Learning for Action Recognition
取得了现在轻量级模型最好的结果
本文提出渐进式增强模块,用于 channel-level 信息滤波,有效地激发了不同帧的鉴别通道,同时避免了重复信息提取。
另外,提出一个时序多样性损失来训练网络。该损失可以校准卷积核,从而使网络可以专注于并捕捉帧之间的变化。也提高了识别精度,且不增加额外的网络复杂性。
https://arxiv.org/pdf/2007.07626.pdf
https://zhuanlan.zhihu.com/p/162026102
(2)CVPR2020 中科院+商汤 SmallBigNet
SmallBigNet: Integrating Core and Contextual Views for Video Classification
模型更加精简,最后得到的模型大小与2D CNN相近,FLOPs翻倍,在Kinetics-400、Something-Something V1&V2上都超过了最近的一些方法。
https://arxiv.org/pdf/2006.14582v1.pdf
https://zhuanlan.zhihu.com/p/153471137
https://github.com/xhl-video/SmallBigNet
(代码还在整理,还没放出来)
(3)CVPR2019 Kaiming He Non-local Neural Networks
Non-local Neural Networks
convolution和recurrent都是对局部区域进行的操作,所以它们是典型的local operations。受计算机视觉中经典的非局部均值(non-local means)的启发,本文提出一种non-local operations用于捕获长距离依赖(long-range dependencies)
https://arxiv.org/pdf/1711.07971v1.pdf
https://github.com/facebookresearch/video-nonlocal-net
https://blog.csdn.net/elaine_bao/article/details/80821306
https://www.zhihu.com/question/68473183