Fig. 4From: Consistent constraint-based video-level learning for action recognitionVideo-level loss functionBack to article page