A 3D CNN(Convolution neural network) implementation in Keras for action recognition.This model trained on videos. This model architecture achieved 96% accuracy after some hours of training on my GPU(RTX 2080TI).
This model architecture and the prediction scripts are such like that, It works on a video and in real time also. Now free to play around the model, change numbers of layers, size, data.
Create a directory "action_recognition_dataset/data" and put all your downloaded in the data folder. And put all your test data in "action_recognition_dataset" directory.
- keras
- glob
- opencv
- numpy