GraSP-VL: Length as a Semantic Granularity Interface for Vision-Language Representations

TL;DR

GraSP-VL introduces a method to turn fixed-length vision-language embeddings into a controllable semantic interface using shared prefix transforms, enabling multi-resolution semantic access without retraining.

Contribution

It proposes GraSP-VL, a novel shared prefix transform that reorganizes frozen embeddings into a hierarchical semantic interface with controllable granularity.

Findings

Achieves high staircase score and hard-negative selectivity on COCO/Flickr30K.

Transfers effectively to SugarCrepe-clean with high object accuracy.

Maintains full-dimensional zero-shot CIFAR-100 accuracy.

Abstract

Frozen vision-language embeddings contain signals at multiple semantic resolutions, from object identity to attributes, relations, and full-caption meaning, but they expose these signals through a fixed-length vector interface. We study whether embedding length can be turned into a controllable semantic access interface. We propose \textbf{GraSP-VL}, which learns a shared near-orthogonal prefix transform over frozen VLM embeddings. GraSP-VL instantiates a \textbf{Semantic Matryoshka} interface: short prefixes are assigned coarse semantic roles, while longer prefixes progressively expose finer language-grounded distinctions. Because the transform is shared across image and text embeddings and preserves full-dimensional geometry, prefix behavior changes without rewriting the original VLM space. On a 20,147-example COCO/Flickr30K annotation pool, GraSP-VL reaches a staircase score of 53.01 and hard-negative selectivity of 89.76, while keeping full-space drift below $10^{-6}$. It also transfers to SugarCrepe-clean with 86.03 object accuracy and 11.96 mean external emergence, and preserves full-dimensional zero-shot CIFAR-100 accuracy. These results show that frozen VLM embeddings can be reorganized into a truncatable semantic prefix interface rather than merely compressed.