Fig. 5From: Consistent constraint-based video-level learning for action recognitionRecognition accuracy with different weight αBack to article page