Datasets:

BDRC
/

tibetan-script-classification-benchmark

language:
  - bo
license: mit
task_categories:
  - image-classification
tags:
  - tibetan
  - manuscript
  - script-classification
  - benchmark
  - bdrc
pretty_name: Tibetan Script Classification Benchmark
size_categories:
  - n<1K
dataset_info:
  features:
    - name: id
      dtype: string
    - name: image_bytes
      dtype: image
    - name: script
      dtype:
        class_label:
          names:
            '0': Danyig
            '1': Druma
            '2': Gyuyig
            '3': Pedri
            '4': Tsugdri
            '5': Uchen
            '6': multiscript
            '7': non_tibetan
    - name: script_type
      dtype: string
  splits:
    - name: test
      num_bytes: 0
      num_examples: 480
  download_size: 0
  dataset_size: 0
configs:
  - config_name: default
    data_files:
      - split: test
        path: test.parquet

Tibetan Script Classification Benchmark

Holdout benchmark for 6-class Tibetan script classification. Test split only — not used during training. All images are BDRC manuscript page scans, balanced by subclass.

Class	Images	Subclasses
Danyig	60	DraDring: 25, DraRing: 9, Drathung: 17, Gongshabma: 3, Tsegdrig: 6
Druma	60	Dhumri: 22, DruDring: 20, DruRing: 10, Druchen: 2, Druthung: 6
Gyuyig	60	Khyuyig: 31, Tsumachug: 15, Yigchung: 14
Pedri	60	Peri: 44, Petsuk: 16
Tsugdri	60	Trinyig: 36, Tsugchung: 14, Tsugthung: 10
Uchen	60	Uchen SugDring: 53, Uchen SugRing: 3, Uchen Sugthung: 4
multiscript	60	Multi-Scripts: 60
non_tibetan	60	Other: 60

Total: 480 images across 8 classes.

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: `Danyig`, `Druma`, `Gyuyig`, `Pedri`, `Tsugdri`, `Uchen`, `multiscript`, `non_tibetan`
`script_type`	string	Sub-script / subclass name (e.g. `DraDring`, `Multi-Scripts`, `Other`)

Load in Python

from datasets import load_dataset

ds = load_dataset("BDRC/tibetan-script-classification-benchmark", split="test")
print(len(ds))  # 480

row = ds[0]
# row["id"], row["image_bytes"], row["script"]

Evaluate a model

from experiments.benchmark_eval.eval import run_benchmark
run_benchmark(model, repo_id="BDRC/tibetan-script-classification-benchmark")

Citation

@misc{bdrcscriptbenchmark,
  title  = {Tibetan Script Classification Benchmark},
  author = {Buddhist Digital Resource Center and OpenPecha},
  year   = {2026},
  url    = {https://huggingface.co/datasets/BDRC/tibetan-script-classification-benchmark},
  note   = {Images from BDRC. MIT.}
}

Acknowledgements

Images from the Buddhist Digital Resource Center (BDRC). Developed by Dharmaduta / OpenPecha.