Scaling Vision Models Does Not Consistently Improve Localisation-Based Explanation Quality

2026-05-11

AI Summary

This study investigates whether scaling up vision models consistently improves the localization quality of post-hoc explanations. The authors systematically evaluate 11 ResNet, DenseNet, and Vision Transformer models of varying sizes across three datasets with ground-truth segmentation masks, employing five prominent explanation methods—including Grad-CAM and Integrated Gradients—and assessing performance using Relevance Rank Accuracy alongside a newly proposed Dual-Polarity Precision metric. Their findings reveal, for the first time, that model scale does not exhibit a consistent positive correlation with explanation quality: higher predictive accuracy does not guarantee better localization, smaller models often outperform larger ones, and some highly accurate models demonstrate localization capabilities barely above random chance.

Abstract

Artificial intelligence models are increasingly scaled to improve predictive accuracy, yet it remains unclear whether scale improves the quality of post-hoc explanations. We investigate this relationship by evaluating 11 computer vision models representing increasing levels of depth and complexity within the ResNet, DenseNet, and Vision Transformer families, trained from scratch or pretrained, across three image datasets with ground-truth segmentation masks. For each model, we generate explanations using five post-hoc explainable AI methods and quantify mask alignment using two localisation metrics: Relevance Rank Accuracy (Arras et al., 2022) and the proposed Dual-Polarity Precision, which measures positive attributions inside the class mask and negative attributions outside it. Across datasets and methods, increasing architectural depth and parameter count does not improve explanation quality in most statistical comparisons, and smaller models often match or exceed deeper variants. While pretraining typically improves predictive performance and increases the dependence of explanations on learned weights, it does not consistently increase localisation scores. We also observe scenarios in which models achieve strong predictive performance while localisation precision is near zero, suggesting that performance metrics alone may not indicate whether predictions are based on the annotated regions. These results indicate that larger models do not reliably provide higher-quality explanations, and that explainability should therefore be assessed explicitly during model selection for safety-sensitive deployments.

Problem

• model scaling
• explainable AI
• localisation-based explanation
• vision models
• post-hoc explanation

Innovation

• model scaling
• post-hoc explanation
• localisation-based explanation
• Dual-Polarity Precision
• explainable AI