Improving Multi-modal Large Language Model through Boosting Vision Capabilities

TL;DR

This paper introduces Arcana, a multimodal language model that enhances visual understanding by employing specialized modules for better integration and learning from visual and textual data, leading to improved multimodal performance.

Contribution

The paper proposes MM-LoRA and QLadder, novel techniques for disentangled multimodal learning and visual feature enhancement in large language models.

Findings

Arcana outperforms baseline models in visual understanding tasks.

The proposed modules significantly improve multimodal integration.

Extensive experiments validate the effectiveness and generalization of Arcana.

Abstract

We focus on improving the visual understanding capability for boosting the vision-language models. We propose \textbf{Arcana}, a multiModal language model, which introduces two crucial techniques. First, we present Multimodal LoRA (MM-LoRA), a module designed to enhance the decoder. Unlike traditional language-driven decoders, MM-LoRA consists of two parallel LoRAs -- one for vision and one for language -- each with its own parameters. This disentangled parameters design allows for more specialized learning in each modality and better integration of multimodal information. Second, we introduce the Query Ladder adapter (QLadder) to improve the visual encoder. QLadder employs a learnable ``\textit{ladder}'' structure to deeply aggregates the intermediate representations from the frozen pretrained visual encoder (e.g., CLIP image encoder). This enables the model to learn new and informative visual features, as well as remaining the powerful capabilities of the pretrained visual encoder. These techniques collectively enhance Arcana's visual perception power, enabling it to leverage improved visual information for more accurate and contextually relevant outputs across various multimodal scenarios. Extensive experiments and ablation studies demonstrate the effectiveness and generalization capability of our Arcana. The code and re-annotated data are available at \url{https://arcana-project-page.github.io}.