README.md

---
license: cc0-1.0
task_categories:
  - object-detection
language:
  - bo
tags:
  - yolo
  - tibetan
  - document-layout-analysis
  - bounding-box
pretty_name: TDLA Training Dataset
size_categories:
  - 1K<n<10K
---

# TDLA Training Dataset

YOLO-format object-detection dataset for **Tibetan Document Layout Analysis (TDLA)**. The dataset contains bounding-box annotations for four layout classes found in Tibetan document page images and is split into training and validation sets using iterative multi-label stratification.

## Overview

| Property | Value |
|---|---|
| **Total images** | 5588 |
| **Total annotations** | 13826 |
| **Number of classes** | 4 |
| **Image format** | JPEG (`.jpg`) |
| **Label format** | YOLO (`.txt`) |
| **Split ratio** | 80% train / 20% val |
| **Stratification** | Iterative multi-label stratification |
| **Random seed** | 42 |

## Image Source

All images in this dataset are sourced from the [Buddhist Digital Resource Center (BDRC)](https://www.bdrc.io/) digital library.

## Classes

| ID | Name | Images | % of dataset |
|---|---|---|---|
| 0 | header | 4280 | 76.6% |
| 1 | Text area | 5532 | 99.0% |
| 2 | footnote | 374 | 6.7% |
| 3 | footer | 3640 | 65.1% |

## Annotation Process

Annotations were created on the [Ultralytics HUB](https://hub.ultralytics.com/) platform using the following 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 dedicated reviewer inspected each annotated image, verifying label correctness, box tightness, and class assignment before the annotation was accepted into the dataset.

## Split Methodology

The dataset was split into **80% training / 20% validation** using **iterative multi-label stratification** (seed = 42). This approach treats each image as a multi-label sample (an image may contain several classes simultaneously) and iteratively assigns images to splits so that per-class proportions stay as close to the target ratio as possible. The result is a near-uniform 80/20 distribution for every class, as shown in the tables below.

## Split Statistics

| Split | Images | % of total |
|---|---|---|
| train | 4470 | 80.0% |
| val | 1118 | 20.0% |

## Class Distribution per Split

| Class | train | val | Total |
|---|---|---|---|
| header | 3424 (80.0%) | 856 (20.0%) | 4280 |
| Text area | 4425 (80.0%) | 1107 (20.0%) | 5532 |
| footnote | 299 (79.9%) | 75 (20.1%) | 374 |
| footer | 2912 (80.0%) | 728 (20.0%) | 3640 |

## Directory Structure

```
TDLA_Training_dataset/
├── images/
│   ├── train/
│   └── val/
├── labels/
│   ├── train/
│   └── val/
├── train.txt
├── val.txt
├── data.yaml
└── README.md
```

## Usage

Point your YOLO training config to `data.yaml` in this directory:

```bash
yolo detect train data=TDLA_Training_dataset/data.yaml
```

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

## Label Format

Each `.txt` label file uses the 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.

## License

This dataset is released under the [CC0 1.0 Universal (Public Domain Dedication)](https://creativecommons.org/publicdomain/zero/1.0/). You are free to copy, modify, and distribute the data, even for commercial purposes, without asking permission.

## Acknowledgements

This dataset was developed by [Dharmaduta](https://dharmaduta.org/) from specifications provided by the [Buddhist Digital Resource Center (BDRC)](https://www.bdrc.io/) for the BDRC Etext Corpus, with funding from the [Khyentse Foundation](https://khyentsefoundation.org/).