Reproduce “Multiple Document Datasets Pre-training Improves Text Line Detection With Deep Neural Networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mélodie Boillet, Christopher Kermorvant, Thierry Paquet
Doc-UFCN detects text lines in historical document images using pixel-wise labeling and dilated convolutions that provide wider page context. It is trained from scratch, while comparisons with ImageNet-pretrained dhSegment vary in both architecture and initialization. On Balsac, its text-line F1 is 91.11%, versus 84.81% for dhSegment, while READ-Simple favors dhSegment; the comparison does not isolate initialization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.