Vision Transformers (ViT)

Patch Merging

Operation in hierarchical transformers that combines groups of 2x2 adjacent patches to create lower-resolution tokens, thereby increasing depth and receptive field.

← Terug