Improving Vision Transformers by Overlapping Heads in Multi-Head Self-Attention

ViT
Re-assign community