DenseFusion-1M: Merging Vision Experts for Comprehensive Multimodal Perception

TL;DR

DenseFusion-1M introduces a large, high-quality dataset with dense image descriptions generated by a novel caption engine, significantly enhancing multimodal large language models' ability to understand complex visual elements.

Contribution

The paper presents DenseFusion-1M, a new dataset created with a cost-effective caption engine that improves MLLMs' perception of detailed visual information.

Findings

Outperforms existing caption engines in generating dense, accurate descriptions.

Enhances MLLMs' performance on diverse vision-language benchmarks.

Enables better understanding of high-resolution images in multimodal models.

Abstract

Existing Multimodal Large Language Models (MLLMs) increasingly emphasize complex understanding of various visual elements, including multiple objects, text information, and spatial relations. Their development for comprehensive visual perception hinges on the availability of high-quality image-text datasets that offer diverse visual elements and throughout image descriptions. However, the scarcity of such hyper-detailed datasets currently hinders progress within the MLLM community. The bottleneck stems from the limited perceptual capabilities of current caption engines, which fall short in providing complete and accurate annotations. To facilitate the cutting-edge research of MLLMs on comprehensive vision perception, we thereby propose Perceptual Fusion, using a low-budget but highly effective caption engine for complete and accurate image descriptions. Specifically, Perceptual Fusion integrates diverse perception experts as image priors to provide explicit information on visual elements and adopts an efficient MLLM as a centric pivot to mimic advanced MLLMs' perception abilities. We carefully select 1M highly representative images from uncurated LAION dataset and generate dense descriptions using our engine, dubbed DenseFusion-1M. Extensive experiments validate that our engine outperforms its counterparts, where the resulting dataset significantly improves the perception and cognition abilities of existing MLLMs across diverse vision-language benchmarks, especially with high-resolution images as inputs. The dataset and code are publicly available at https://github.com/baaivision/DenseFusion.