Keep-or-Drop? Adaptive Tokenizer for Compact Video Representation | Yeonkyeong Lee et al. | ResearchPod