README.md

metadata

license: other
license_name: fair-use-no-warranty
license_link: LICENSE
task_categories:
  - object-detection
language:
  - bo
tags:
  - yolo
  - tibetan
  - document-layout-analysis
  - bounding-box
size_categories:
  - 1K<n<10K
pretty_name: TDLA Training Dataset v2
extra_gated_prompt: >-
  The page images in this dataset are scans of Tibetan texts from the BDRC
  digital library and are provided on a FAIR-USE basis for research. No
  copyright license is granted. By requesting access you acknowledge that you
  are solely responsible for performing your own copyright / rights analysis
  before any use, and that the Buddhist Digital Resource Center (BDRC) accepts
  no liability for any misuse of this material.
extra_gated_fields:
  Full name: text
  Affiliation: text
  Intended use: text
  I have read the copyright notice and will perform my own copyright analysis before use: checkbox
  I understand BDRC is not liable for any misuse of this material: checkbox

TDLA Training Dataset v2

YOLO-format object-detection dataset for Tibetan Document Layout Analysis (TDLA). It contains bounding-box annotations for four layout classes on scanned Tibetan document pages, split into training, validation, and test sets.

This is an expanded, re-reviewed successor to BDRC/TDLA-Training-Dataset, built from several annotation batches that were consolidated to a single, consistent annotation convention and split to be leakage-free.

Overview

Property	Value
Total annotations	25460
Total images	8325
Number of classes	4
Image format	JPEG (.jpg)
Label format	YOLO (.txt)
Splits	train / val / test
Split unit	volume-level (leakage-free)

Image Source

All images are sourced from the Buddhist Digital Resource Center (BDRC) digital library.

Classes

ID	Name	Annotations	% of total
0	header	8155	32.0%
1	text-area	10705	42.0%
2	footnote	367	1.4%
3	footer	6233	24.5%

Annotation Process

Annotations were created on the Ultralytics HUB platform in a two-stage workflow:

1. Annotation — annotators drew bounding boxes for each of the four layout classes (header, text-area, footnote, footer) on every page image.
2. Quality control — a reviewer inspected every image, verifying label correctness, box tightness, and class assignment. Earlier annotation batches were re-reviewed so that all sources follow the same convention (in particular, marginal header/footer elements are boxed per element, consistently across the whole dataset).
3. Automated consistency audit — a final geometric/logical audit flagged likely mistakes (near-duplicate or conflicting-class boxes, impossible header/footer/footnote orderings, out-of-bounds boxes). Flagged pages were manually corrected and re-imported, removing conflicting duplicate boxes.

Split Methodology

The train / val / test split is created by grouping pages at the volume (book) level and assigning each volume as a whole to a single split. This guarantees there is no leakage across splits — no page (or an augmented copy of it) and no volume appears in more than one split. The split has been audited for pixel-identical duplicates, shared page identities, and shared volumes across splits (all clean).

• Footnote stratification — the footnote class is rare, so footnote-bearing volumes were distributed across all three splits to keep the class represented everywhere.
• Augmented data — a subset of the training images are augmented (geometric/photometric) copies. These are confined to the training set only; validation and test contain exclusively original, non-augmented scans, making them a clean benchmark. Augmented images can be recognised by an __aug marker in their filename.
• Approximate ratio: ~81% train / ~9% val / ~10% test by image count.

Split Statistics

Split	Images
train	6751
val	714
test	860

(train includes 1197 augmented images; val and test include 0 and 0.)

Annotation Distribution per Split

Class	train	val	test	Total
header	6638	671	846	8155
text-area	8722	858	1125	10705
footnote	296	26	45	367
footer	5046	540	647	6233

A single image can contain multiple annotations of the same class, so annotation counts may exceed image counts.

Directory Structure

TDLA-Training-Dataset-v2/
├── images/
│   ├── train/
│   ├── val/
│   └── test/
├── labels/
│   ├── train/
│   ├── val/
│   └── test/
├── train.txt
├── val.txt
├── test.txt
├── data.yaml
└── README.md

Usage

Point your YOLO training config at data.yaml:

yolo detect train data=data.yaml

The train.txt, val.txt, and test.txt files list relative image paths for each split.

Label Format

Each .txt label file uses standard YOLO format — one row per bounding box:

<class_id> <x_center> <y_center> <width> <height>

All coordinates are normalized to [0, 1] relative to image dimensions.

Copyright & Usage Notice

This dataset does not come with an open-content license. The page images are scans of Tibetan texts from the BDRC digital library and are distributed on a fair-use basis for research and non-commercial layout-analysis work.

• No copyright license is granted over the underlying page images.
• You are solely responsible for performing your own copyright / rights analysis for your jurisdiction and intended use before using this material.
• BDRC accepts no liability for any misuse of this material.

By accessing the gated dataset you accept these terms.

Acknowledgements

Developed by the Buddhist Digital Resource Center (BDRC) for the BDRC Etext Corpus. Thanks to the annotators and reviewers who produced and consolidated the layout annotations.