Fig. 4From: Improved two-stream model for human action recognitionThe proposed structure in the spatial modelBack to article page