A Novel Framework for Automated Explain Vision Model Using Vision-Language Models

2025-08-27

AI Summary

Current vision models prioritize accuracy-oriented metrics (e.g., IoU, mAP) while neglecting interpretability; mainstream XAI methods provide only instance-level attribution explanations and fail to characterize global behavioral patterns across entire datasets, hindering bias detection and failure analysis. Method: We propose the first automated dual-granularity explanation framework grounded in vision-language models (VLMs), unifying instance-level attribution mapping with dataset-level semantic abstraction. Leveraging prompt learning and cross-modal alignment, it generates human-readable, semantically rich behavioral explanations without model fine-tuning. Contribution/Results: Experiments demonstrate that our framework efficiently identifies failure cases, uncovers decision preferences and latent biases, and enhances model transparency and trustworthiness—all at low computational cost. It bridges the gap between local explainability and global behavioral understanding in vision systems.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Humans and AI: Explainable AI (XAI) for Human Understanding
• Machine Learning: Transparent, Interpretable, Explainable ML

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

The development of many vision models mainly focuses on improving their performance using metrics such as accuracy, IoU, and mAP, with less attention to explainability due to the complexity of applying xAI methods to provide a meaningful explanation of trained models. Although many existing xAI methods aim to explain vision models sample-by-sample, methods explaining the general behavior of vision models, which can only be captured after running on a large dataset, are still underexplored. Furthermore, understanding the behavior of vision models on general images can be very important to prevent biased judgments and help identify the model's trends and patterns. With the application of Vision-Language Models, this paper proposes a pipeline to explain vision models at both the sample and dataset levels. The proposed pipeline can be used to discover failure cases and gain insights into vision models with minimal effort, thereby integrating vision model development with xAI analysis to advance image analysis.

Problem

• Explaining vision models' general behavior on large datasets
• Reducing complexity of applying xAI methods for meaningful explanations
• Preventing biased judgments by understanding model trends and patterns

Innovation

• Vision-language models for explainable AI
• Pipeline for sample and dataset level analysis
• Automated failure case discovery and insights