Using tf-openpose as preprocessing, and adding deep learning model to do action recognition.
Before running demo, you should download graph files.
- cmu (trained in 656x368)
- mobilenet_thin (trained in 432x368)
CMU's model graphs are too large for git, so I uploaded them on an external cloud. You should download them if you want to use cmu's original model. Download scripts are provided in the model folder.
$ cd models/graph/cmu
$ bash download.sh
Download labeled video dataset to train the action recognizing model
$ bash download_dataset.sh