visual-token-compression

[NeurIPS 2026 Spotlight] Braco: Beyond Selection: Token Parameterization for Extreme Visual Token Compression. Official training and inference code for efficient vision-language models (VLMs).

AssocVTC: Association-trained Visual Token Compression — train an image encoder to pack a whole scene into few decisive visual tokens for MLLM/embodied-AI context budget. (formerly SR-Diffusion-v3)

Just Respect the Semantics: Image-Token Compression via Pre-LLM Semantic Variable-Resolution Tokenization and In-LLM Relevance Pruning