DEAL: Disentangle and Localize Concept-level Explanations for VLMs
This addresses the issue of poor explainability in VLMs for users needing fine-grained concept understanding, representing an incremental improvement through a novel method for a known bottleneck.
The paper tackles the problem of vision-language models (VLMs) producing entangled and mislocalized explanations for fine-grained concepts, proposing the DEAL method to disentangle and localize these explanations without human annotations, which significantly improves disentanglability and localizability and even boosts prediction accuracy by reducing reliance on spurious correlations.
Large pre-trained Vision-Language Models (VLMs) have become ubiquitous foundational components of other models and downstream tasks. Although powerful, our empirical results reveal that such models might not be able to identify fine-grained concepts. Specifically, the explanations of VLMs with respect to fine-grained concepts are entangled and mislocalized. To address this issue, we propose to DisEntAngle and Localize (DEAL) the concept-level explanations for VLMs without human annotations. The key idea is encouraging the concept-level explanations to be distinct while maintaining consistency with category-level explanations. We conduct extensive experiments and ablation studies on a wide range of benchmark datasets and vision-language models. Our empirical results demonstrate that the proposed method significantly improves the concept-level explanations of the model in terms of disentanglability and localizability. Surprisingly, the improved explainability alleviates the model's reliance on spurious correlations, which further benefits the prediction accuracy.