Braco：面向极端视觉 token 压缩的 token 参数化方法

View PDF HTML (experimental)

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.

Comments:	Accepted at NeurIPS 2026 (Spotlight). Code: this https URL
Subjects:	Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Cite as:	arXiv:2609.35232 [cs.CV]
	(or arXiv:2609.35232v2 [cs.CV] for this version)
	https://doi.org/10.48550/arXiv.2609.35232 arXiv-issued DOI via DataCite

Submission history

From: Rui Zhong [view email]
[v1] Mon, 28 Sep 2026 14:10:25 UTC (1,923 KB)
[v2] Tue, 29 Sep 2026 14:19:02 UTC (1,921 KB)