[VISAPP 2026] Graph-Guided Token Merging (G2TM) is a lightweight one-shot module designed to eliminate redundant tokens in the early layers of a ViT-based models, through graph theory. It performs a single token merging step after a shallow attention block, enabling all subsequent layers to operate on a compact token set.
machine-learning deep-learning neural-network graph pytorch artificial-intelligence transformer networkx token segmenter vision-transformer token-reduction token-merging g2tm
-
Updated
Apr 10, 2026 - Python