ADEM-VL: Adaptive and Embedded Fusion for Efficient Vision-Language Tuning

TL;DR

ADEM-VL is proposed, an efficient vision-language method that tunes VL models based on pretrained large language models by adopting a parameter-free cross-attention mechanism for similarity measurements in multimodal fusion that outperforms existing approaches with reduced training and inference latency.

Abstract

Recent advancements in multimodal fusion have witnessed the remarkable success of vision-language (VL) models, which excel in various multimodal applications such as image captioning and visual question answering. However, building VL models requires substantial hardware resources, where efficiency is restricted by two key factors: the extended input sequence of the language model with vision features demands more computational operations, and a large number of additional learnable parameters increase memory complexity. These challenges significantly restrict the broader applicability of such models. To bridge this gap, we propose ADEM-VL, an efficient vision-language method that tunes VL models based on pretrained large language models (LLMs) by adopting a parameter-free cross-attention mechanism for similarity measurements in multimodal fusion. This approach only requires embedding vision features into the language space, significantly reducing the number of trainable parameters and accelerating both training and inference speeds. To enhance representation learning in fusion module, we introduce an efficient multiscale feature generation scheme that requires only a single forward pass through the vision encoder. Moreover, we propose an adaptive fusion scheme that dynamically discards less relevant visual information for each text token based on its attention score. This ensures that the fusion process prioritizes the most pertinent visual features. With experiments on various tasks including visual question answering, image captioning, and instruction-following, we demonstrate that our framework outperforms existing approaches. Specifically, our method surpasses existing methods by an average accuracy of 0.77% on ScienceQA dataset, with reduced training and inference latency, demonstrating the superiority of our framework. The code is available at https://github.com/Hao840/ADEM-VL.

Figures (4)

• Figure 1: Comparison of different vision-language tuning frameworks: (a) Methods that directly extend the input space of the language model with extracted vision features. (b) Methods that fuse vision information into the language model via cross-attention. (c) Our proposed ADEM-VL framework, which incorporates parameter-free cross-attention, multiscale visual prompting, and adaptive multimodal fusion designs. This approach ensures both parameter and computational efficiency while delivering promising performance.
• Figure 2: Comparison of different hyperparameter settings in the ADEM-VL with LLaMA-7B as the language model.
• Figure 3: Visualization of image captioning results with LLaMA-7B. In each row, the left figure is the original image, while the middle and right figures demonstrate the dropping decisions for features at two different scales.
• Figure 4: Examples of zero-shot instruction-following tasks with LLaMA-7B.