MatCha: Enhancing Visual Language Pretraining with Math Reasoning and Chart Derendering

TL;DR

MatCha significantly improves visual language models' ability to understand and reason about charts and plots by introducing specialized pretraining tasks, leading to substantial performance gains on benchmark datasets and better transfer to diverse visual data.

Contribution

The paper introduces MatCha, a novel pretraining approach that enhances visual language models with math reasoning and chart derendering capabilities, improving performance on chart-related tasks.

Findings

MatCha outperforms previous methods by up to 20% on PlotQA and ChartQA benchmarks.

Pretraining with MatCha improves transfer to diverse visual domains like screenshots and diagrams.

MatCha demonstrates the effectiveness of specialized pretraining tasks for visual language understanding.

Abstract

Visual language data such as plots, charts, and infographics are ubiquitous in the human world. However, state-of-the-art vision-language models do not perform well on these data. We propose MatCha (Math reasoning and Chart derendering pretraining) to enhance visual language models' capabilities in jointly modeling charts/plots and language data. Specifically, we propose several pretraining tasks that cover plot deconstruction and numerical reasoning which are the key capabilities in visual language modeling. We perform the MatCha pretraining starting from Pix2Struct, a recently proposed image-to-text visual language model. On standard benchmarks such as PlotQA and ChartQA, the MatCha model outperforms state-of-the-art methods by as much as nearly 20%. We also examine how well MatCha pretraining transfers to domains such as screenshots, textbook diagrams, and document figures and observe overall improvement, verifying the usefulness of MatCha pretraining on broader visual language tasks.