Memvit: Memory-augmented multiscale vision transformer for efficient long-term video recognition

摘要: … The techniques presented in this paper are general and applicable to other transformer-based video models. We hope MeMViT will be useful for future long-term video modeling …