search

XiaoCode-er / Two-Stream-CNN

Two Stream CNN implemented in Keras using in skeleton-based action recognition with dataset NTU RGB+D
76 stars 13 forks source link

Multi-person Question #2

Closed dihuangcode closed 6 years ago

dihuangcode commented 6 years ago

Thanks for your amazing work! Here comes one question: in NTU, some videos contain only one person, but some may contain two or more persons. So I am confused about how you solve this. I read your code, which shows you only solve two-person situation. Am I right?

XiaoCode-er commented 6 years ago

If some videos contain only one person, you can construct another person with padding zero. This operating had processed in hdf5 file. This model is shared for every input，so extract the feature from conv part and then fuse multi-person feature.

dihuangcode commented 6 years ago

thanks a lot! @XiaoCode-er I find in some videos(e.g. S015C001P025R002A059_rgb.avi), the body count is wrong(should be 2 but 3). How do you solve this problem?

XiaoCode-er commented 6 years ago

@Yedis You can refer to the link https://github.com/InwoongLee/TS-LSTM and you will know how to process this problem.

dihuangcode commented 6 years ago

Thanks a lot!