DreamStruct: Understanding Slides and User Interfaces via Synthetic Data Generation

TL;DR

DreamStruct introduces a synthetic data generation method for structured visuals like slides and UIs, enabling effective machine understanding with minimal manual annotation, thereby improving recognition, description, and classification tasks.

Contribution

It presents a novel code-based synthetic data generation approach that reduces manual labeling and enhances model performance on structured visual understanding tasks.

Findings

Improved recognition accuracy for visual elements.

Enhanced content description capabilities.

Better classification of visual content types.

Abstract

Enabling machines to understand structured visuals like slides and user interfaces is essential for making them accessible to people with disabilities. However, achieving such understanding computationally has required manual data collection and annotation, which is time-consuming and labor-intensive. To overcome this challenge, we present a method to generate synthetic, structured visuals with target labels using code generation. Our method allows people to create datasets with built-in labels and train models with a small number of human-annotated examples. We demonstrate performance improvements in three tasks for understanding slides and UIs: recognizing visual elements, describing visual content, and classifying visual content types.