English

PreSTU: Pre-Training for Scene-Text Understanding

Computer Vision and Pattern Recognition 2023-08-22 v3 Computation and Language

Abstract

The ability to recognize and reason about text embedded in visual inputs is often lacking in vision-and-language (V&L) models, perhaps because V&L pre-training methods have often failed to include such an ability in their training objective. In this paper, we propose PreSTU, a novel pre-training recipe dedicated to scene-text understanding (STU). PreSTU introduces OCR-aware pre-training objectives that encourage the model to recognize text from an image and connect it to the rest of the image content. We implement PreSTU using a simple transformer-based encoder-decoder architecture, combined with large-scale image-text datasets with scene text obtained from an off-the-shelf OCR system. We empirically demonstrate the effectiveness of this pre-training approach on eight visual question answering and four image captioning benchmarks.

Keywords

Cite

@article{arxiv.2209.05534,
  title  = {PreSTU: Pre-Training for Scene-Text Understanding},
  author = {Jihyung Kil and Soravit Changpinyo and Xi Chen and Hexiang Hu and Sebastian Goodman and Wei-Lun Chao and Radu Soricut},
  journal= {arXiv preprint arXiv:2209.05534},
  year   = {2023}
}

Comments

Accepted to ICCV 2023

R2 v1 2026-06-28T01:09:40.696Z