Bridging Compressed Image Latents and Multimodal Large Language Models

TL;DR

This paper introduces a lightweight framework to adapt compressed image latents for multimodal large language models, enabling efficient vision tasks on resource-constrained devices without retraining entire MLLMs.

Contribution

We propose a novel, general framework that adapts neural image compression outputs for MLLMs, excluding the large MLLM components from training, which is more practical and scalable.

Findings

Achieves high rate-accuracy performance with less complexity.

Compatible with various neural image codecs and MLLMs.

Effective across multiple application scenarios.

Abstract

This paper presents the first-ever study of adapting compressed image latents to suit the needs of downstream vision tasks that adopt Multimodal Large Language Models (MLLMs). MLLMs have extended the success of large language models to modalities (e.g. images) beyond text, but their billion scale hinders deployment on resource-constrained end devices. While cloud-hosted MLLMs could be available, transmitting raw, uncompressed images captured by end devices to the cloud requires an efficient image compression system. To address this, we focus on emerging neural image compression and propose a novel framework with a lightweight transform-neck and a surrogate loss to adapt compressed image latents for MLLM-based vision tasks. Given the huge scale of MLLMs, our framework excludes the entire downstream MLLM except part of its visual encoder from training our system. This stands out from most existing coding for machine approaches that involve downstream networks in training and thus could be impractical when the networks are MLLMs. The proposed framework is general in that it is applicable to various MLLMs, neural image codecs, and multiple application scenarios, where the neural image codec can be (1) pre-trained for human perception without updating, (2) fully updated for joint human and machine perception, or (3) fully updated for only machine perception. Extensive experiments on different neural image codecs and various MLLMs show that our method achieves great rate-accuracy performance with much less complexity.