RaVL: Discovering and Mitigating Spurious Correlations in Fine-Tuned Vision-Language Models
This addresses robustness issues in vision-language models for applications like image classification, though it is incremental as it builds on existing spurious correlation mitigation methods by focusing on fine-grained features.
The paper tackles the problem of spurious correlations in fine-tuned vision-language models, which degrade zero-shot performance, by proposing RaVL to discover and mitigate these correlations using local image features, resulting in a 191% improvement in discovery accuracy and an 8.2% improvement in worst-group classification accuracy.
Fine-tuned vision-language models (VLMs) often capture spurious correlations between image features and textual attributes, resulting in degraded zero-shot performance at test time. Existing approaches for addressing spurious correlations (i) primarily operate at the global image-level rather than intervening directly on fine-grained image features and (ii) are predominantly designed for unimodal settings. In this work, we present RaVL, which takes a fine-grained perspective on VLM robustness by discovering and mitigating spurious correlations using local image features rather than operating at the global image level. Given a fine-tuned VLM, RaVL first discovers spurious correlations by leveraging a region-level clustering approach to identify precise image features contributing to zero-shot classification errors. Then, RaVL mitigates the identified spurious correlation with a novel region-aware loss function that enables the VLM to focus on relevant regions and ignore spurious relationships during fine-tuning. We evaluate RaVL on 654 VLMs with various model architectures, data domains, and learned spurious correlations. Our results show that RaVL accurately discovers (191% improvement over the closest baseline) and mitigates (8.2% improvement on worst-group image classification accuracy) spurious correlations. Qualitative evaluations on general-domain and medical-domain VLMs confirm our findings.