Acceleration

The article presents AiluRus, a new method to speed up Vision Transformers (ViTs) for dense prediction tasks like object detection and semantic segmentation. The approach reduces the complexity of ViTs…