Vision-Language Models (VLMs) achieved strong performance on a variety of tasks (e.g., image-text retrieval, visual question answering). However, most VLMs rely on coarse-grained image-caption pairs for alignment, relying on data volume to resolve ambiguities and ground linguistic concepts in images. The richer semantic and syntactic structure within text is largely overlooked. To address this, we propose HIerarchically STructured Learning (HIST) that enhances VLM training without any additional supervision, by hierarchically decomposing captions into the constituent Subject, Noun Phrases, and Composite Phrases. Entailment between these constituent components allows us to formulate additional regularization constraints on the VLM attention maps. Specifically, we introduce two novel loss functions: (1) Subject Loss, which aligns image content with the subject of corresponding phrase, acting as an entailment of standard contrastive/matching losses at the Phrase level; (2) Addition Loss, to balance attention across multiple objects. HIST is general, and can be applied to any VLM for which attention between vision and language can be computed; we illustrate its efficacy on BLIP and ALBEF. HIST outperforms baseline VLMs, achieving up to +9.8% improvement in visual grounding, +6.3% in multi-object referring segmentation, +1.1% in image-text retrieval, and +0.2% in visual question answering, underscoring the value of structuring learning in VLMs.

Recent advances in pixel-level tasks (e.g., segmentation) illustrate the benefit of long-range interactions between aggregated region-based representations that can enhance local features. However, such pixel-to-region associations and the resulting representation, which often take the form of attention, cannot model the underlying semantic structure of the scene (e.g., individual objects and, by extension, their interactions). In this work, we take a step toward addressing this limitation. Specifically, we propose an architecture where we learn to project image features into latent region representations and perform global reasoning across them, using a transformer, to produce contextualized and scene-consistent representations that are then fused with original pixel-level features. Our design enables the latent regions to represent semantically meaningful concepts, by ensuring that activated regions are spatially disjoint and unions of such regions correspond to connected object segments. The resulting semantic global reasoning (SGR) is end-to-end trainable and can be combined with any semantic segmentation framework and backbone. Combining SGR with DeepLabV3 results in a semantic segmentation performance that is competitive to the state-of-the-art, while resulting in more semantically interpretable and diverse region representations, which we show can effectively transfer to detection and instance segmentation. Further, we propose a new metric that allows us to measure the semantics of representations at both the object class and instance level.