Explaining Vision GNNs: A Semantic and Visual Analysis of Graph-based Image Classification

2025-04-28

AI Summary

This paper addresses the limited interpretability of Graph Neural Network (GNN)-based vision models in image classification. We systematically investigate the semantic consistency and spatial coherence across multiple graph layers to assess how well these models capture object structure and spatial relationships, and we compare explanation patterns between clean and adversarial samples. We propose the first quantitative framework for evaluating cross-layer graph connectivity, integrating semantic similarity metrics, spatial coherence analysis, and heatmap-guided information flow visualization. Experiments reveal that while GNN vision models exhibit baseline interpretability, their deeper-layer reasoning substantially diverges from human visual perception. Our framework effectively uncovers such intrinsic biases, providing both theoretical foundations and a reproducible evaluation paradigm for developing trustworthy, cognitively aligned visual GNNs. (136 words)

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Graph Neural Networks (GNNs) have emerged as an efficient alternative to convolutional approaches for vision tasks such as image classification, leveraging patch-based representations instead of raw pixels. These methods construct graphs where image patches serve as nodes, and edges are established based on patch similarity or classification relevance. Despite their efficiency, the explainability of GNN-based vision models remains underexplored, even though graphs are naturally interpretable. In this work, we analyze the semantic consistency of the graphs formed at different layers of GNN-based image classifiers, focusing on how well they preserve object structures and meaningful relationships. A comprehensive analysis is presented by quantifying the extent to which inter-layer graph connections reflect semantic similarity and spatial coherence. Explanations from standard and adversarial settings are also compared to assess whether they reflect the classifiers' robustness. Additionally, we visualize the flow of information across layers through heatmap-based visualization techniques, thereby highlighting the models' explainability. Our findings demonstrate that the decision-making processes of these models can be effectively explained, while also revealing that their reasoning does not necessarily align with human perception, especially in deeper layers.

Problem

• Analyzing semantic consistency in GNN-based image classification graphs
• Quantifying inter-layer graph connections for semantic and spatial coherence
• Assessing explainability and robustness in adversarial settings

Innovation

• Patch-based graph construction for image classification
• Semantic and spatial analysis of GNN layers
• Heatmap visualization for model explainability