arXiv Computation and Language

When Modality Gap Reduction Fails: Prediction-Level Hubness in CLIP

The paper investigates why reducing the modality gap between image and text representations in CLIP does not always improve zero‑shot classification accuracy. It shows that while average alignment improves, the relative decision margins among classes can shift, leading to a prediction‑level hubness where predictions concentrate on a few classes. Experiments across datasets confirm that accuracy drops correlate with increased prediction concentration for both linear and learning‑based gap corrections.

arXiv AI
Sep 30

One Geometry, Different Outcomes: Readout-Dependent Effects of the Modality Gap in Vision-Language Models

The paper investigates how the modality gap— the separation between image and text representations in contrastive vision‑language models—affects different downstream tasks. By showing that a single dominant direction accounts for most of the image‑text mean separation, the authors explain why reducing or removing this gap can improve zero‑shot classification, degrade retrieval, or restore performance depending on the task. The study provides a geometric framework that clarifies when and why gap interventions should be applied in vision‑language systems.

By Aditya Sharma, Divya Saxena
arXiv Computer Vision
Oct 1

Project and Mix: Task-Semantic Prototypes for Few-Shot Image Classification

arXiv:2603.24528v2 Announce Type: replace Abstract: Vision-language models like CLIP are trained with the objective of aligning text and image pairs. Beyond text prompts alone, recent works show that...

By Dipam Goswami, Simone Magistri, Gido M. van de Ven, Bart{\l}omiej Twardowski, Andrew D. Bagdanov, Tinne Tuytelaars, Joost van de Weijer
arXiv AI
Jun 8

Modality Gap-Driven Subspace Alignment Training Paradigm For Multimodal Large Language Models

arXiv:2602. 07026v3 Announce Type: replace-cross Abstract: Despite the success of multimodal contrastive learning in aligning visual and linguistic representations, a persistent geometric anomaly, the Modality Gap, remains: embeddings of distinct modalities expressing identical semantics occupy systematically offset regions.

By Xiaomin Yu, Yi Xin, Yuhui Zhang, Wenjie Zhang, Chonghan Liu, Hanzhen Zhao, Chen Liu, Xiaoxing Hu, Ziyue Qiao, Hao Tang, Xiaobin Hu, Chengwei Qin, Hui Xiong, Yu Qiao, Shuicheng Yan
arXiv Computer Vision
Oct 16

Flow-CLIP: Image-Conditioned Residual Scoring for CompositionalImage-Text Matching

Flow-CLIP introduces a method to improve compositional image-text matching by refining global image-text alignment with image-conditioned changes in token–patch evidence. It combines Global Compatibility Calibration (GCC) for sentence-level similarity, Image-Conditioned Evidence Refinement (ICER) to incorporate visual patch context, and Cross-Image Margin Ranking (CIMR) to enhance ranking margins when images support correct descriptions. Experiments demonstrate that Flow-CLIP outperforms pretrained CLIP and other global alignment approaches on binding accuracy and reduces reliance on language priors.

By Minghong Wei, Enguang Zhu, Pu Cao, Lu Yang, Qing Song
arXiv Computer Vision
Oct 9

Image Recognition with Vision and Language Embeddings of VLMs

The paper evaluates how vision-language models (VLMs) perform in image classification tasks, comparing language-guided and vision-only approaches across various dual-encoder VLMs such as SigLIP 2 and RADIOv2.5 on ImageNet-1k and its label-corrected variant. It identifies key factors influencing accuracy—including prompt design, class diversity, k‑NN neighbors, and reference set size—and demonstrates that language and vision provide complementary strengths. A simple, learning‑free fusion method based on per‑class precision is proposed to further improve classification performance.

By Illia Volkov, Nikita Kisel, Klara Janouskova, Jiri Matas
arXiv AI
Aug 28

Subspace Alignment for Vision-Language Model Test-time Adaptation

The paper introduces SubTTA, a test-time adaptation method for vision‑language models that aligns the semantic subspaces of visual and textual modalities to improve zero‑shot predictions. It addresses two issues: the modality gap caused by distribution shifts and visual nuisance that masks task‑specific semantics. By minimizing chordal distance between principal subspaces and projecting visual features onto a task‑specific textual subspace, SubTTA refines decision boundaries and achieves an average 2.24% improvement over existing TTA methods.

By Zhichen Zeng, Wenxuan Bao, Xiao Lin, Ruizhong Qiu, Tianxin Wei, Xuying Ning, Yuchen Yan, Chen Luo, Monica Xiao Cheng, Jingrui He, Hanghang Tong