Abstract
Visual-token compression is effective for improving the efficiency of vision-language models, but under extreme compression budgets, token pruning can break visual grounding while learned resamplers increase parameter count, attention cost, and training complexity. We revisit compression through a token parameterization lens, separating (i) basis transformation and structured truncation (retained subspace/compressibility) from (ii) coordinate organization (optimization and cross-modal alignment). This view yields two coupled objectives, compressibility and learnability, which we formalize as unified functionals. Guided by these objectives, we design Braco, a lightweight four-step coder that combines transform-basis truncation, input-independent basis-coordinate embeddings, budget-dependent orthogonal re-parameterization, and learned spatial residual tokens from lightweight pooling. Experiments show that Braco forms the favorable empirical accuracy-efficiency frontier under $23\times$--$64\times$ compression and remains competitive at $144\times$, reaching 95.2% accuracy while reducing prefill FLOPs by 84.2%--86.7% relative to the uncompressed upper bound. Against prior methods, Braco matches or improves accuracy while achieving up to approximately 36% end-to-end speedup and using $16.6\times$/$78.8\times$ lower compressor latency/FLOPs.
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Zhong, R., Li, Y., Yan, Z., & Zhuo, C. (2026). Beyond Selection: Token Parameterization for Extreme Visual Token Compression. https://omanscience.com/en/articles/beyond-selection-token-parameterization-for-extreme-visual-token-compression
MLA 9
Zhong, Rui, et al. "Beyond Selection: Token Parameterization for Extreme Visual Token Compression." https://omanscience.com/en/articles/beyond-selection-token-parameterization-for-extreme-visual-token-compression.
Chicago (author–date)
Zhong, Rui, Yu Li, Zheyu Yan, and Cheng Zhuo. 2026. "Beyond Selection: Token Parameterization for Extreme Visual Token Compression." https://omanscience.com/en/articles/beyond-selection-token-parameterization-for-extreme-visual-token-compression.
Harvard
Zhong, R., Li, Y., Yan, Z. and Zhuo, C. (2026) 'Beyond Selection: Token Parameterization for Extreme Visual Token Compression', Available at: https://omanscience.com/en/articles/beyond-selection-token-parameterization-for-extreme-visual-token-compression.
Vancouver
Zhong R, Li Y, Yan Z, Zhuo C. Beyond Selection: Token Parameterization for Extreme Visual Token Compression. https://omanscience.com/en/articles/beyond-selection-token-parameterization-for-extreme-visual-token-compression
IEEE
R. Zhong, Y. Li, Z. Yan, and C. Zhuo, "Beyond Selection: Token Parameterization for Extreme Visual Token Compression," https://omanscience.com/en/articles/beyond-selection-token-parameterization-for-extreme-visual-token-compression.