Long Story Short: Disentangling Compositionality and Long-Caption Understanding in VLMs
This addresses a key limitation in vision-language models for applications requiring detailed image-text alignment, though it is incremental in improving existing capabilities.
The paper tackled the challenge of understanding long, dense captions in vision-language models by investigating the relationship with compositionality, finding that training for one enhances the other, with models achieving strong performance in both tasks when using high-quality data.
Contrastive vision-language models (VLMs) have made significant progress in binding visual and textual information, but understanding long, dense captions remains an open challenge. We hypothesize that compositionality, the capacity to reason about object-attribute bindings and inter-object relationships, is key to understanding longer captions. In this paper, we investigate the interaction between compositionality and long-caption understanding, asking whether training for one property enhances the other. We train and evaluate a range of models that target each of these capabilities. Our results reveal a bidirectional relationship: compositional training improves performance on long-caption retrieval, and training on long captions promotes compositionality. However, these gains are sensitive to data quality and model design. We find that training on poorly structured captions, or with limited parameter updates, fails to support generalization. Likewise, strategies that aim at retaining general alignment, such as freezing positional embeddings, do not improve compositional understanding. Overall, we find that compositional understanding and long-caption understanding are intertwined capabilities that can be jointly learned through training on dense, grounded descriptions. Despite these challenges, we show that models trained on high-quality, long-caption data can achieve strong performance in both tasks, offering practical guidance for improving VLM generalization.