ECCV VISART 26ECCV: CCF BECCV: ICORE A*First Author
Full-Page Chinese Calligraphy Character Detection: A Dataset and Benchmark with Integrity-Preserving Tiling and SDAC-Det
T. Liu, N. van Noord, C. Liang, K. Zhang
ECCV 2026 VISART VIII (to appear), 2026
Abstract
Full-page Chinese calligraphy character detection is challenging because artworks vary in layout, scale, and resolution. We introduce a benchmark of 2,144 pages and 189,156 character boxes from 13 source subsets, with in-domain and held-out-domain evaluation. Integrity-preserving tiling avoids ambiguous supervision from truncated characters, while SDAC-Det combines local tile features with full-page density and scale priors. The benchmark provides a foundation for analyzing calligraphy beyond standard OCR.
calligraphydeep-learningcultural-heritage
BibTeX
@inproceedings{Liu2026SDACDet,
author = {Liu, Tiancheng and van Noord, Nanne and Liang, Chen and Zhang, Kang},
title = {Full-Page Chinese Calligraphy Character Detection: A Dataset and Benchmark with Integrity-Preserving Tiling and {SDAC-Det}},
booktitle = {VISART VIII: Vision for Art and Culture Workshop, European Conference on Computer Vision (ECCV)},
year = {2026},
note = {Accepted; to appear}
}