Datasets:

hydroshiba
/

hcmus-doc-layout

text string
6 0.314715 0.101981 0.106027 0.016462
8 0.529018 0.128239 0.189732 0.020307
12 0.395647 0.154576 0.264509 0.021205
8 0.516741 0.24721 0.508929 0.155134
8 0.516741 0.486607 0.508929 0.314732
8 0.516741 0.766183 0.508929 0.235491
5 0.516183 0.914062 0.052455 0.017857
6 0.314715 0.101981 0.106027 0.016462
8 0.515919 0.197078 0.510572 0.155134
8 0.516741 0.474888 0.508929 0.33817
8 0.516741 0.766183 0.508929 0.235491
5 0.516183 0.914062 0.052455 0.017857
13 0.339286 0.288504 0.151786 0.018973
6 0.314715 0.101981 0.106027 0.016462
13 0.313058 0.128906 0.09933 0.018973
8 0.50558 0.154576 0.419643 0.021205
14 0.449777 0.261719 0.308036 0.179687
8 0.516741 0.499442 0.508929 0.235491
8 0.516741 0.726004 0.508929 0.208705
5 0.516183 0.914062 0.052455 0.017857
13 0.435268 0.366071 0.34375 0.022321
11 0.319754 0.106585 0.11942 0.02567
11 0.347656 0.127232 0.175223 0.022321
8 0.516741 0.193638 0.508929 0.101562
8 0.516741 0.366071 0.508929 0.234375
11 0.316406 0.501674 0.106027 0.018973
8 0.516741 0.581473 0.508929 0.125
8 0.532924 0.659598 0.460938 0.022321
4 0.518415 0.77567 0.487723 0.207589
0 0.516741 0.887835 0.337054 0.018973
11 0.516183 0.914062 0.052455 0.017857
6 0.315278 0.102451 0.106003 0.017403
13 0.3125 0.128762 0.100446 0.018685
8 0.516741 0.207031 0.508929 0.128348
7 0.516741 0.301897 0.508929 0.045759
8 0.445871 0.446429 0.364955 0.022321
8 0.488839 0.474888 0.386161 0.018973
8 0.439732 0.501674 0.287946 0.018973
8 0.516741 0.566406 0.508929 0.101562
8 0.516741 0.699777 0.508929 0.154018
5 0.516183 0.913504 0.052455 0.018973
7 0.516741 0.353286 0.508929 0.045759
7 0.516183 0.407021 0.507813 0.045759
6 0.31529 0.102633 0.106027 0.017766
7 0.516741 0.154018 0.513393 0.071429
8 0.516741 0.313616 0.508929 0.073661
8 0.516741 0.539621 0.508929 0.36942
8 0.516741 0.766183 0.508929 0.074777
8 0.455915 0.820871 0.320312 0.018973
8 0.516741 0.859933 0.508929 0.047991
5 0.516183 0.914062 0.059152 0.017857
7 0.516741 0.232701 0.508929 0.074777
6 0.31529 0.102283 0.106027 0.017066
8 0.516741 0.28683 0.513393 0.341518
8 0.517299 0.618862 0.510045 0.313616
5 0.516183 0.914062 0.059152 0.017857
6 0.31529 0.101672 0.106027 0.015843
14 0.516741 0.253259 0.511161 0.274732
12 0.470982 0.397879 0.415179 0.021205
8 0.516741 0.489955 0.508929 0.15625
7 0.516741 0.665179 0.508929 0.178571
7 0.516741 0.823661 0.508929 0.127232
5 0.516741 0.913504 0.060268 0.018973
11 0.55692 0.127232 0.279018 0.022321
12 0.421875 0.15904 0.232143 0.021205
8 0.525112 0.24442 0.52567 0.127232
8 0.525112 0.384487 0.52567 0.126116
8 0.525112 0.536272 0.52567 0.152902
8 0.525112 0.662388 0.52567 0.072545
8 0.525112 0.746094 0.52567 0.077009
8 0.469866 0.805804 0.328125 0.022321
7 0.550964 0.866256 0.492553 0.074777
5 0.545759 0.939732 0.011161 0.011161
7 0.521205 0.12779 0.433036 0.021205
8 0.470424 0.371652 0.331473 0.022321
7 0.489955 0.403648 0.370536 0.023814
12 0.535156 0.566964 0.460938 0.022321
8 0.525112 0.625 0.52567 0.075893
13 0.340402 0.683594 0.069196 0.016741
8 0.525112 0.727679 0.52567 0.049107
7 0.525112 0.799107 0.52567 0.075893
5 0.546317 0.939732 0.010045 0.015625
7 0.546317 0.180246 0.483259 0.074777
7 0.545015 0.248049 0.483259 0.045205
7 0.545666 0.313338 0.483259 0.070312
7 0.546317 0.457589 0.483259 0.071429
7 0.546317 0.522879 0.483259 0.047991
7 0.525112 0.866071 0.52567 0.049107
7 0.525112 0.141741 0.52567 0.049107
0 0.481585 0.189174 0.429688 0.021205
14 0.525112 0.33817 0.527902 0.258929
8 0.397879 0.504464 0.186384 0.022321
8 0.525112 0.563058 0.52567 0.074777
8 0.525112 0.660714 0.52567 0.102679
5 0.546317 0.94029 0.012277 0.014509
4 0.546317 0.271205 0.481027 0.314732
0 0.546875 0.450335 0.241071 0.018973
8 0.525112 0.521205 0.52567 0.102679
13 0.34096 0.59375 0.070312 0.015625
8 0.525112 0.665179 0.52567 0.102679

End of preview.

HCMUS Document-Layout Detection

Fine-grained document-layout object detection on thesis pages: real scanned HCMUS (Vietnamese) theses plus programmatically generated synthetic pages.

Layout (mirrored under the repo root)

Path	Contents
`images/train/shard00` … `shard03`	26,988 JPEG (20,292 unique pages: 6,696 human-reviewed real + 13,596 synthetic; each real page replayed twice as `rep2_*` aliases → effective ~1:1 real:synthetic per epoch)
`images/val/`	916 JPEG, real only, held out at the author/document level
`annotations/train/shard00` … `shard03`	YOLO txt, same shard/filename as the images
`annotations/val/`	YOLO txt
`train.json`, `val.json`	COCO annotations

Directories are sharded because the Hub allows ≤ 10,000 files per directory.

Classes (15)

cap, catalogue, code, equ, fig, foot, header, list, para, para_equ, reference, sec_1, sec_2, sec_3, tab

Formats — both annotation kinds provided

COCO train.json / val.json — images / annotations / categories. Category id = line index in the class list above (0-based); boxes in absolute pixels (x y w h); iscrowd: 0. file_name is repo-relative, e.g. images/train/shard00/HCMUS_2020_..._p001.jpg.
YOLO txt under annotations/... — one line per box, same filename as the image, class_id cx cy w h normalized.

Images are 896×896.

Real vs synthetic

Real: scanned human-reviewed HCMUS theses, split by author so a thesis never straddles train/val (shared fonts/margins/header style would otherwise leak near-duplicates across the split).
Synthetic: LaTeX-generated pages targeting two corpus weaknesses — severe class imbalance (e.g. code is ~0.2% of boxes) and single-genre narrowness. Boxes are exact by construction.

Notes

The validation set is deliberately real-only: a metric measured partly on generated pages would report how well the model learned the generator rather than how well it reads documents.
Training (and the reference detector) used this train set as-is, replay included — rep2_* entries duplicate real pages and are not novel data.
test/ and benchmark splits are intentionally not published here.

Downloads last month: 35

Total file size:

3.2 GB