From Local Details to Global Context: Advancing Vision-Language Models with Attention-Based Selection

Lincan Cai, Jingxuan Kang, Shuang Li, Wenxuan Ma, Binhui Xie, Zhida Qin, Jian Liang

TL;DR: We propose a training-free method that enhances vision-language model performance by using attention maps for fine-grained raw space selection and holistic feature selection.

Abstract: Pretrained vision-language models (VLMs), e.g., CLIP, demonstrate impressive zero-shot capabilities on downstream tasks. Prior research highlights the crucial role of visual augmentation techniques, like random cropping, in alignment with fine-grained class descriptions generated by large language models (LLMs), significantly enhancing zero-shot performance by incorporating multi-view information. However, the inherent randomness of these augmentations can inevitably introduce background artifacts and cause models to overly focus on local details, compromising global semantic understanding. To address these issues, we propose an **A**ttention-**B**ased **S**election (**ABS**) method from local details to global context, which applies attention-guided cropping in both raw images and feature space, supplement global semantic information through strategic feature selection. Additionally, we introduce a soft matching technique to effectively filter LLM descriptions for better alignment. **ABS** achieves state-of-the-art performance on out-of-distribution generalization and zero-shot classification tasks. Notably, **ABS** is training-free and even rivals few-shot and test-time adaptation methods.

Lay Summary: we propose ABS (Attention-Based Selection), a method that smartly selects the most important parts of an image, both in the original pixels and in the model’s internal features, which can align better with the detailed LLM descriptions of each image. This enhances the zero-shot classification performance of vision-language models.

Link To Code: https://github.com/BIT-DA/ABS

Primary Area: Deep Learning->Other Representation Learning

Keywords: Vision-language models, Attention-Based, Feature selection

Submission Number: 4487