README.md

---
language:
  - bo
task_categories:
  - image-classification
tags:
  - tibetan
  - manuscript
  - script-classification
  - dinov3
  - bdrc
pretty_name: 4-Class Tibetan Script Classification
size_categories:
  - 1K<n<10K
dataset_info:
  features:
    - name: id
      dtype: string
    - name: image_bytes
      dtype: image
    - name: script
      dtype:
        class_label:
          names:
            '0': Druma
            '1': Uchen
            '2': Danyig+Pedri
            '3': Gyuyig+Tsugdri
  splits:
    - name: train
      num_bytes: 0
      num_examples: 1920
    - name: validation
      num_bytes: 0
      num_examples: 240
    - name: test
      num_bytes: 0
      num_examples: 240
  download_size: 0
  dataset_size: 0
configs:
  - config_name: default
    data_files:
      - split: train
        path: "train/train-*.parquet"
      - split: validation
        path: "val/val-*.parquet"
      - split: test
        path: "test/test-*.parquet"
---

# 4-Class Tibetan Script Classification

Manuscript script classification on BDRC-style page images (4 classes).

This release uses **uniform per-class sampling** (600 images per class across all splits combined).

## Images per class

| Class | train | val | test | **All** |
|-------|------:|----:|-----:|--------:|
| Druma | 480 | 60 | 60 | 600 |
| Uchen | 480 | 60 | 60 | 600 |
| Danyig+Pedri | 480 | 60 | 60 | 600 |
| Gyuyig+Tsugdri | 480 | 60 | 60 | 600 |

## Splits

| Split | Images | 
|-------|-------:|
| train | 1,920 | 
| validation | 240 |
| test | 240 | 
| **Total** | **2,400** |

Page-level split manifest: [`splits/splits.json`](splits/splits.json).

## Parquet schema

| Column | Type | Description |
|--------|------|-------------|
| `id` | string | BDRC page id (e.g. `W00KG09391-I00KG093950005`) |
| `image_bytes` | binary | JPEG/PNG page image |
| `script` | string | One of: `Druma`, `Uchen`, `Danyig+Pedri`, `Gyuyig+Tsugdri` |

Shards: `train/train-*.parquet`, `val/val-*.parquet`, `test/test-*.parquet`.

See [`split_stats.json`](split_stats.json) and [`split_stats.md`](split_stats.md) for row-level counts.

## Load in Python

```python
from datasets import load_dataset

ds = load_dataset("BDRC/4-class-tibetan-script-classification-dataset")
train = ds["train"]           # 1,920
val   = ds["validation"]      # 240
test  = ds["test"]            # 240
```

```python
from io import BytesIO
from PIL import Image

row = train[0]
img = Image.open(BytesIO(row["image_bytes"])).convert("RGB")
print(row["id"], row["script"])
```

## Train a model

```bash
python scripts/upload_dataset.py --repo-id BDRC/4-class-tibetan-script-classification-dataset
```

## Citation

```bibtex
@misc{bdrcscriptclass,
  title  = {4-Class Tibetan Script Classification Dataset},
  author = {Buddhist Digital Resource Center and OpenPecha},
  year   = {2026},
  url    = {https://huggingface.co/datasets/BDRC/4-class-tibetan-script-classification-dataset},
  note   = {Images from BDRC}
}
```

## License

Images taken from the open access collection of the Buddhist Digital Resource Center. Not all images are in the public domain, some are from recent publications possibly under copyright. We provide the images under the Fair Use copyright exception, but any reuse of this dataset will have to be based on a copyright analysis. We provide the classification data under the CC0 1.0 Universal (Public Domain Dedication).

## Acknowledgements

All images are provided by the Buddhist Digital Resource Center (BDRC). This dataset was developed by Dharmaduta from specifications provided by BDRC for the project "The BDRC Etext Corpus", with funding from the Khyentse Foundation. **[Buddhist Digital Resource Center](https://www.bdrc.io)** (BDRC). Developed by Dharmaduta / OpenPecha.