MATE: Meet At The Embedding -- Connecting Images with Long Texts
This addresses a limitation in VLMs for handling complex text interactions with long texts, offering a solution for applications requiring image-document alignment, though it is incremental as it builds on existing VLM and LLM technologies.
The paper tackles the problem of aligning images with long texts, which is underexplored in Vision Language Models, by introducing MATE, a method that integrates a pretrained LLM-based encoder with a VLM and a projection module, achieving effective cross-modal retrieval on new benchmarks.
While advancements in Vision Language Models (VLMs) have significantly improved the alignment of visual and textual data, these models primarily focus on aligning images with short descriptive captions. This focus limits their ability to handle complex text interactions, particularly with longer texts such as lengthy captions or documents, which have not been extensively explored yet. In this paper, we introduce Meet At The Embedding (MATE), a novel approach that combines the capabilities of VLMs with Large Language Models (LLMs) to overcome this challenge without the need for additional image-long text pairs. Specifically, we replace the text encoder of the VLM with a pretrained LLM-based encoder that excels in understanding long texts. To bridge the gap between VLM and LLM, MATE incorporates a projection module that is trained in a multi-stage manner. It starts by aligning the embeddings from the VLM text encoder with those from the LLM using extensive text pairs. This module is then employed to seamlessly align image embeddings closely with LLM embeddings. We propose two new cross-modal retrieval benchmarks to assess the task of connecting images with long texts (lengthy captions / documents). Extensive experimental results demonstrate that MATE effectively connects images with long texts, uncovering diverse semantic relationships.