![](https://img-blog.csdnimg.cn/20201014180756928.png?x-oss-process=image/resize,m_fixed,h_64,w_64)
行为识别
文章平均质量分 62
Alex丶Chen
这个作者很懒,什么都没留下…
展开
-
mmaction2 指南 (7) 自定义运行配置
mmaction2 指南Outlinemmaction2 指南7. 自定义Runtime Settings使用pytorch的优化器自定义优化器附加配置自定义训练策略自定义工作流 Workflow自定义 Hooks1. 实现新的hook2. 注册3. 修改配置使用MMCV定义的hooks修改 checkpoint hook修改 log hook修改 evaluation hook7. 自定义Runtime Settings使用pytorch的优化器optimizer = dict(type='Ada原创 2021-01-27 23:35:12 · 953 阅读 · 0 评论 -
mmaction2 指南 (5) 自定义新模块
mmaction2 指南Outlinemmaction2 指南5. 自定义新模块自定义优化器自定义优化器 Constructor自定义新组件新的backbone新的head新的损失添加新的学习策略(learning rate scheduler (updater))5. 自定义新模块自定义优化器参考 CopyOfSGD例子:mmaction/core/optimizer/my_optimizer.pyfrom .registry import OPTIMIZERSfrom torch.opt原创 2021-01-27 23:09:41 · 959 阅读 · 0 评论 -
mmaction2 指南 (6) 导出模型到ONNX
Open Neural Network Exchange (ONNX) is an open ecosystem that empowers AI developers to choose the right tools as their project evolves.mmaction2 指南Outlinemmaction2 指南6. 导出模型到ONNX6. 导出模型到ONNX支持:I3DTSNTINTSMR(2+1)DSLOWFASTSLOWONLYBMNBSN(tem,.原创 2021-01-27 23:05:59 · 813 阅读 · 0 评论 -
mmaction2 指南 (4) 自定义数据处理流程
mmaction2 指南Outlinemmaction2 指南4. 自定义数据处理流程流程设计lazy opsExtend and Use Custom Pipelines 自定义一个pipeline4. 自定义数据处理流程流程设计RawframeDataset 原始数据SampleFrames 帧采样(原视频切clip,clip长度,帧间隔,clip数量,【后续流程继承 frame_inds 帧index】)FrameSelector(【后续流程继承imgs 图片列表】,原尺寸)Resize原创 2021-01-27 21:21:05 · 1260 阅读 · 1 评论 -
mmaction2 指南 (3) 添加新的数据集
mmaction2 指南Outlinemmaction2 指南3. 添加新的数据集自定义行为识别数据集新的数据重新组织到已有格式自定义数据集的例子支持混合数据训练3. 添加新的数据集自定义行为识别数据集新的数据重新组织到已有格式重新构建标注文件就行标注格式一:rawframe标注【路径,帧数,类别】some/directory-1 163 1some/directory-2 122 1some/directory-3 258 2some/directory-4 234 2some原创 2021-01-27 16:23:31 · 1377 阅读 · 0 评论 -
mmaction2 指南 (2) Finetune模型
mmaction2 指南Outlinemmaction2 指南2. Finetunning 模型修改任务头 Head修改数据集修改训练策略使用pre-trained模型2. Finetunning 模型Step 1: 定义数据集Step 2: 修改配置修改任务头 Head2D/3D模型,backbone配置,分类头/定位头/检测头的配置(TSN,TSM,BSN)model = dict( type='Recognizer2D', backbone=dict( t原创 2021-01-27 15:24:05 · 699 阅读 · 0 评论 -
mmaction2 指南 (1) 配置文件
mmaction2 指南Outlinemmaction2 指南1. 配置文件配置文件命名规则配置文件structureAction localizartion 配置Action Recognition 配置配置文件中的中间变量1. 配置文件https://github.com/open-mmlab/mmaction2/blob/master/docs/tutorials/1_config.md配置文件命名规则{model}_[model setting]_{backbone}_[misc]_{da原创 2021-01-27 15:23:45 · 1125 阅读 · 0 评论 -
mmaction2 sthv2 数据处理
参考 自己走一遍STEP 1下载数据,解压到videos路径:220,847 个后缀为webm的视频STEP 2cd $MMACTION2/tools/data/sthv2/bash extract_rgb_frames.sh#!/usr/bin/env bashcd ../python build_rawframes.py ../../data/sthv2/videos/ ../../data/sthv2/rawframes/ --task rgb --level 1 --ext原创 2021-01-26 19:35:52 · 664 阅读 · 0 评论 -
行为识别论文笔记|SlowFast|SlowFast Networks for Video Recognition
行为识别论文笔记|SlowFast|SlowFast Networks for Video RecognitionFeichtenhofer, Christoph, et al. “Slowfast networks for video recognition.” Proceedings of the IEEE international conference on computer vision. 2019. ——FAIRMotivationsThe categorical spatial se原创 2020-12-07 12:56:21 · 727 阅读 · 0 评论 -
行为识别论文笔记|TSM|TSM: Temporal Shift Module for Efficient Video Understanding
行为识别论文笔记|TSM|TSM: Temporal Shift Module for Efficient Video UnderstandingLin, Ji , C. Gan , and S. Han . “TSM: Temporal Shift Module for Efficient Video Understanding.” 2019 IEEE/CVF International Conference on Computer Vision (ICCV) IEEE, 2019.Motiv原创 2020-12-07 12:49:37 · 2089 阅读 · 0 评论 -
行为识别论文笔记|Non-local|Non-local Neural Networks
行为识别论文笔记-Non-Local-Non-local Neural NetworksWang, Xiaolong, et al. “Non-local neural networks.” Proceedings of the IEEE conference on computer vision and pattern recognition. 2018. 引用数2000+ cmu(Wangxiaolong)和FAIR(HeKaiming)MotivationConv and recurre原创 2020-12-07 12:41:08 · 530 阅读 · 0 评论 -
行为识别论文笔记|SlowFusion|Large-scale Video Classification with Convolutional Neural Networks|
Large-scale Video Classification with Convolutional Neural Networks 论文笔记Andrej Karpathy et al. “Large-scale Video Classification with Convolutional Neural Networks” CVPR 2014Motivation缺少使用CNN进行视频分类的数据集和方法,所以推出了数据集 Sports-1M(1 million YouTube videos,原创 2020-09-24 16:42:10 · 929 阅读 · 0 评论 -
行为识别论文笔记|TSN|Temporal Segment Networks: Towards Good Practices for Deep Action Recognition
Temporal Segment Networks: Towards Good Practices for Deep Action Recognition 论文笔记Temporal Segment Networks: Towards Good Practices for Deep Action Recognition, Limin Wang, Yuanjun Xiong, Zhe Wang, Yu Qiao, Dahua Lin, Xiaoou Tang, and Luc Van Gool, ECCV原创 2020-09-25 19:32:01 · 842 阅读 · 0 评论 -
行为识别论文笔记|I3D S3D R(2+1)D P3D CSN
视频理解论文笔记-I3D S3D R(2+1)D P3D CSNI3DCarreira, Joao, and Andrew Zisserman. “Quo vadis, action recognition? a new model and the kinetics dataset.” proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017.S3DXie, Saining, et al.原创 2020-11-30 08:45:50 · 3946 阅读 · 0 评论 -
行为识别论文笔记|ARTNet|Appearance-and-Relation Networks for Video Classification
视频理解论文笔记-ARTNet-Appearance-and-Relation Networks for Video ClassificationWang, Limin, et al. “Appearance-and-relation networks for video classification.” Proceedings of the IEEE conference on computer vision and pattern recognition. 2018.Motivation3 k原创 2020-11-30 08:48:15 · 368 阅读 · 0 评论