Not Only Text: Exploring Compositionality of Visual Representations in Vision-Language Models
This work addresses the challenge of understanding compositional reasoning in VLMs for the vision community, making their processes more interpretable, but it is incremental as it builds on prior studies of compositionality in text representations.
The paper tackled the problem of whether compositional patterns emerge in the visual embedding space of Vision-Language Models (VLMs), and found that visual embeddings exhibit a compositional arrangement, with their proposed Geodesically Decomposable Embeddings (GDE) framework achieving stronger performance in compositional classification and higher results in group robustness compared to baseline methods.
Vision-Language Models (VLMs) learn a shared feature space for text and images, enabling the comparison of inputs of different modalities. While prior works demonstrated that VLMs organize natural language representations into regular structures encoding composite meanings, it remains unclear if compositional patterns also emerge in the visual embedding space. In this work, we investigate compositionality in the image domain, where the analysis of compositional properties is challenged by noise and sparsity of visual data. We address these problems and propose a framework, called Geodesically Decomposable Embeddings (GDE), that approximates image representations with geometry-aware compositional structures in the latent space. We demonstrate that visual embeddings of pre-trained VLMs exhibit a compositional arrangement, and evaluate the effectiveness of this property in the tasks of compositional classification and group robustness. GDE achieves stronger performance in compositional classification compared to its counterpart method that assumes linear geometry of the latent space. Notably, it is particularly effective for group robustness, where we achieve higher results than task-specific solutions. Our results indicate that VLMs can automatically develop a human-like form of compositional reasoning in the visual domain, making their underlying processes more interpretable. Code is available at https://github.com/BerasiDavide/vlm_image_compositionality.