SynthVLM: Towards High-Quality and Efficient Synthesis of Image-Caption Datasets for Vision-Language Models

TL;DR

SynthVLM introduces a novel method for generating high-quality image-caption datasets using diffusion models, leading to improved vision-language models that outperform existing datasets and models on multiple benchmarks.

Contribution

The paper presents SynthVLM, a new data synthesis approach and dataset that enhance vision-language model training and performance, with state-of-the-art results on VQA and MMLU benchmarks.

Findings

SynthVLM-100K dataset outperforms traditional datasets in quality.

SynthVLM-based models achieve SOTA on VQA tasks.

Models trained with SynthVLM data outperform LLaVA with less pretraining data.

Abstract

Vision-Language Models (VLMs) have recently emerged, demonstrating remarkable vision-understanding capabilities. However, training these models requires large-scale datasets, which brings challenges related to efficiency, effectiveness, and quality of web data. In this paper, we introduce SynthVLM, a new data synthesis and curation method for generating image-caption pairs. Unlike traditional methods, where captions are generated from images, SynthVLM utilizes advanced diffusion models and high-quality captions to synthesize and select images from text captions, thereby creating precisely aligned image-text pairs. We further introduce SynthVLM-100K, a high-quality dataset consisting of 100K curated and synthesized image-caption pairs. In both model and human evaluations, SynthVLM-100K outperforms traditional real-world datasets. Leveraging this dataset, we develop a new family of multimodal large language models (MLLMs), SynthVLM-7B and SynthVLM-13B, which achieve state-of-the-art (SOTA) performance on various vision question-answering (VQA) tasks. Notably, our models outperform LLaVA across most metrics with only 18\% pretrain data. Furthermore, SynthVLM-7B and SynthVLM-13B attain SOTA performance on the MMLU benchmark, demonstrating that the high-quality SynthVLM-100K dataset preserves language abilities.