Decoder-Agnostic Token Merging for Vision Transformers: A Systematic Study of G2TM
arXiv CS.CV
•
Generative AI
Computer Vision
AI Research
Vision Transformers (ViTs) have achieved state-of-the-art performance across a range of computer vision tasks, mainly thanks to the self-attention mechanism. We extend Graph-Guided Token Merging (G2TM), a single module inserted early in a ViT-based network, beyond its original Segmenter setting.