🤖 AI Summary
Existing zero-shot vision-language understanding methods rely on global image-text matching, neglecting fine-grained semantic alignments (e.g., object–keyword correspondences), which limits interpretability and generalization. This work proposes the first fine-grained cross-modal alignment framework tailored for zero-shot scenarios, jointly modeling semantic correspondences between region-level visual objects and word-level textual units—without task-specific annotations. Built upon the CLIP backbone, our method integrates region-word attention alignment, contrastive learning optimization, and a multi-task joint zero-shot adaptation mechanism. It achieves significant improvements over state-of-the-art zero-shot approaches on VQA, SNLI-VE, and VCR. Ablation studies confirm the effectiveness of fine-grained modeling and its cross-task transferability, enhancing both reasoning transparency and out-of-distribution generalization.
📝 Abstract
Vision-language tasks, such as VQA, SNLI-VE, and VCR are challenging because they require the model's reasoning ability to understand the semantics of the visual world and natural language. Supervised methods working for vision-language tasks have been well-studied. However, solving these tasks in a zero-shot setting is less explored. Since Contrastive Language-Image Pre-training (CLIP) has shown remarkable zero-shot performance on image-text matching, previous works utilized its strong zero-shot ability by converting vision-language tasks into an image-text matching problem, and they mainly consider global-level matching (e.g., the whole image or sentence). However, we find visual and textual fine-grained information, e.g., keywords in the sentence and objects in the image, can be fairly informative for semantics understanding. Inspired by this, we propose a unified framework to take advantage of the fine-grained information for zero-shot vision-language learning, covering multiple tasks such as VQA, SNLI-VE, and VCR. Our experiments show that our framework outperforms former zero-shot methods on VQA and achieves substantial improvement on SNLI-VE and VCR. Furthermore, our ablation studies confirm the effectiveness and generalizability of our proposed method.
