Get Started
Home
Topics
Search
Library
Research questionHow should synthetic OCR supervision transfer to real printed and handwritten Thai documents without OCR labels?Synthetic pages provide exact labels at scale, but transfer depends on which aspects of real documents are reproduced. Typography, spatial structure, handwriting glyphs, source domain, and page-versus-crop training can affect printed and handwritten recognition differently.
AI
Computer Vision
Evaluation & Benchmarks
Image & Video Processing
Machine Learning
Natural Language Processing
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.How Far Can Synthetic Data Take Thai OCR?The evidence concerns Thai OCR trained on 45,723 synthetic pages and evaluated on real printed and handwritten documents. It includes controlled comparisons of source domain, page context, typography, two-dimensional structure, glyph variation, and page- versus crop-level training, as well as Wayu-Paxa-OCR-Zero adapted from PaddleOCR-VL-1.6 without OCR labels from real pages.research paper · Sep 3, 2026
Related questions
How can vision-language models faithfully transcribe corrupted document text instead of rewriting it?How can OCR diagnose and correct case-level errors while preserving rendering-equivalent outputs?How can low-resource Thai TTS learn a fixed voice from synthetic speech while preserving pronunciation and prosody?How can we generate realistic synthetic identity documents for verification evaluation when real documents are scarce?