Real-Time-Action-Recognition icon indicating copy to clipboard operation
Real-Time-Action-Recognition copied to clipboard

how to train a new model using my own video, what is the input data input format?

Open Jorigorn opened this issue 6 years ago • 6 comments

Hello, nice work, thanks for sharing.

May I ask how can I train a new action recognition model that can recognize somebody is smoking?somebody is writing on the blackboard and so on. Could you provide your training code and annotated video sample to have look at the input format. Thanks a lot. :)

John

Jorigorn avatar Jul 19 '18 18:07 Jorigorn

@Jorigorn Hi,

我英语比较槽糕,直接用中文回答啦。 首先,我这个项目里的行为识别部分没有用任何模型,就是依据关节点的变化规律手动建立了几个行为模板,然后进行动作匹配的(就是凑的)。比如站立的时候,连续的一段时间内关节点是相对比较稳定,水平方向不会有较大的位移,同样,垂直方向也没有较大的位移,我就是这么凑的。

我目前也在想用何种模型来进行行为识别建模。

第二,关于你说的想识别一个人抽烟、写字等动作,我个人的想法是:姿态信息与交互信息的结合。

TianzhongSong avatar Jul 22 '18 00:07 TianzhongSong

谢谢回复。啊,原来如此,你的project在搜索结果上特别靠前。非常有意思的demo. 最近在做行为识别的研究,可能是基于openpose 或者tf-openpose上做。迟点我再做些调研。 再次感谢。

Jorigorn avatar Jul 23 '18 18:07 Jorigorn

近在做行为识别的研究,可能是基于openpose 或者

I am doing the same research:)

jeffzout avatar Dec 06 '18 02:12 jeffzout

@TianzhongSong Hi, I want to add some more features like punching kick etc it would be help for me if you guide the procedure for this Thank you in advance :)

burhanuddin78 avatar Mar 06 '20 06:03 burhanuddin78

您好!请问现在有模型用来训练识别各种行为吗? 期待回复!感谢!

QueenW1998 avatar Dec 07 '21 11:12 QueenW1998

@QueenW1998 Check it out https://github.com/felixchenfy/Realtime-Action-Recognition

burhanuddin78 avatar Dec 09 '21 12:12 burhanuddin78