SyntheticDoc dataset boosts quality of document photo fixes
SyntheticDoc: A Large Synthetic Dataset for Document Unwarping and Illumination Correction
Computer Vision and Pattern RecognitionGraphics
Summary
Photos of documents often look warped or have uneven lighting, making them hard to read or process. The authors created SyntheticDoc, a huge collection of high-quality synthetic images with exact detail to help computers learn to fix these problems better. They made the images with realistic physics and lighting to mimic real photos closely. They showed that training on this new set improves computer models for straightening and lighting correction compared to previous smaller datasets.
What this means in practice
- •For mobile app developers: Improve photo scanning apps by training models with SyntheticDoc to better straighten and light-correct images of documents.
- •For digital archiving teams: Use SyntheticDoc to enhance automated tools that clean up photographs of documents for clearer digital storage.
Authors
Daniel Woortmann, Tanguy Magne, Olga Sorkine-Hornung
Abstract
Deep learning models have become the standard tool for document rectification and illumination correction, yet their performance is fundamentally bound by their training data. For nearly a decade, the community has heavily relied on Doc3D, a pioneering but increasingly limited document unwarping dataset in terms of scale and quality. To address this bottleneck, we introduce SyntheticDoc, a massive, high-quality dataset designed to push the boundaries of document unwarping. SyntheticDoc is composed of 1,000,000 high-resolution procedurally generated training samples, alongside extensive validation and test sets. Each sample is paired with rich, pixel-perfect annotations, including UV maps, normal maps, albedo and shading. To ensure physical accuracy and photorealism, the paper geometries are generated via a physics-based simulator and rendered using a path tracer. To demonstrate the benefit of our dataset, we train a simple baseline model on SyntheticDoc and report on its performance in comparison to state-of-the-art methods on both document unwarping and illumination correction tasks. Our dataset is available at https://igl.ethz.ch/projects/SyntheticDoc/ and the code used to generate it at https://github.com/tanguymagne/SyntheticDoc .