Fig. 2From: Consistent constraint-based video-level learning for action recognitionThe process of clip generationBack to article page