Beyond Heuristic Prompting: A Concept-Guided Bayesian Framework for Zero-Shot Image Recognition

TL;DR

This paper introduces a Bayesian framework for zero-shot image recognition that leverages class-specific concepts, improving prompt quality and robustness over heuristic methods, and achieves superior performance on benchmark datasets.

Contribution

It proposes a novel Bayesian approach incorporating concept priors and likelihoods, with a multi-stage concept synthesis pipeline and adaptive soft-trim likelihood for enhanced zero-shot classification.

Findings

Outperforms state-of-the-art methods on benchmark datasets

Demonstrates robustness to outlier concepts

Provides theoretical guarantees and risk bounds

Abstract

Vision-Language Models (VLMs), such as CLIP, have significantly advanced zero-shot image recognition. However, their performance remains limited by suboptimal prompt engineering and poor adaptability to target classes. While recent methods attempt to improve prompts through diverse class descriptions, they often rely on heuristic designs, lack versatility, and are vulnerable to outlier prompts. This paper enhances prompt by incorporating class-specific concepts. By treating concepts as latent variables, we rethink zero-shot image classification from a Bayesian perspective, casting prediction as marginalization over the concept space, where each concept is weighted by a prior and a test-image conditioned likelihood. This formulation underscores the importance of both a well-structured concept proposal distribution and the refinement of concept priors. To construct an expressive and efficient proposal distribution, we introduce a multi-stage concept synthesis pipeline driven by LLMs to generate discriminative and compositional concepts, followed by a Determinantal Point Process to enforce diversity. To mitigate the influence of outlier concepts, we propose a training-free, adaptive soft-trim likelihood, which attenuates their impact in a single forward pass. We further provide robustness guarantees and derive multi-class excess risk bounds for our framework. Extensive experiments demonstrate that our method consistently outperforms state-of-the-art approaches, validating its effectiveness in zero-shot image classification. Our code is available at https://github.com/less-and-less-bugs/CGBC.