Fetching the paper…

MeMViT: Memory-Augmented Multiscale Vision Transformer for Efficient Long-Term Video Recognition · Around