Fig. 6From: Improved two-stream model for human action recognitionThe structure of the LSTMBack to article page