Probing the Role of Positional Information in Vision-Language Models
This work addresses a critical issue in multimodal AI by showing that detectable information in models may not be usable cross-modally, which is incremental as it builds on existing VL models.
The study investigated how Vision-Language models use positional information, finding that LXMERT fails to leverage it for image-text matching despite its presence in representations, and proposed strategies like pre-training and contrastive learning that improved model properties but had negligible downstream performance gains.
In most Vision-Language models (VL), the understanding of the image structure is enabled by injecting the position information (PI) about objects in the image. In our case study of LXMERT, a state-of-the-art VL model, we probe the use of the PI in the representation and study its effect on Visual Question Answering. We show that the model is not capable of leveraging the PI for the image-text matching task on a challenge set where only position differs. Yet, our experiments with probing confirm that the PI is indeed present in the representation. We introduce two strategies to tackle this: (i) Positional Information Pre-training and (ii) Contrastive Learning on PI using Cross-Modality Matching. Doing so, the model can correctly classify if images with detailed PI statements match. Additionally to the 2D information from bounding boxes, we introduce the object's depth as new feature for a better object localization in the space. Even though we were able to improve the model properties as defined by our probes, it only has a negligible effect on the downstream performance. Our results thus highlight an important issue of multimodal modeling: the mere presence of information detectable by a probing classifier is not a guarantee that the information is available in a cross-modal setup.