openpecha/tibetan-script-classifier
Image Classification • Updated
The dataset viewer should be available soon. Please retry later.
Model: openpecha/uchen-ume-classifier
Hub: openpecha/uchen-ume-classification-benchmark
| File | Rows | Columns |
|---|---|---|
| train.parquet | 9,110 | id, image_bytes, script |
| validation.parquet | 1,000 | same |
| test.parquet | 851 | same |
| benchmark.parquet | 60 | same |
from datasets import load_dataset
repo = "openpecha/uchen-ume-classification-benchmark"
train = load_dataset(repo, split="train")
bench = load_dataset(repo, split="benchmark")
train.py)
| Split | with_preprocess |
without_preprocess |
|---|---|---|
| train | center crop | none |
| val | center crop | none |
| test | none (full page) | none |
Benchmark inference should use --preprocess none for both models when comparing to the test split. Center-cropping at inference on with_preprocess inflates benchmark scores (~98% acc) because it matches val, not test.
preprocess none)
| Model | Accuracy | Macro-F1 | AUC |
|---|---|---|---|
| without_preprocess | 85.0% | 0.848 | 0.970 |
| with_preprocess | 68.3% | 0.648 | 0.953 |
Test split (867 full pages): without_preprocess 80.7% / 0.708; with_preprocess 56.1% / 0.506.
pip install -r requirements-inference.txt
python inference_uchen_ume.py \
--benchmark-dir benchmark \
--model-repo openpecha/uchen-ume-classifier \
--weights without_preprocess/final_model.pt \
--preprocess none \
--output benchmark_predictions.json
with_preprocess (same as training test: no crop)
python inference_uchen_ume.py \
--benchmark-dir benchmark \
--weights with_preprocess/final_model.pt \
--preprocess none
python inference_uchen_ume.py \
--benchmark-json benchmark/benchmark_holdout.json \
--fetch-urls \
--preprocess none
python inference_uchen_ume.py --image page.jpg --weights without_preprocess/final_model.pt --preprocess none
splits/train_val_test_splits.json — pages + image_urlbenchmark/benchmark_holdout.json — 60 holdout pages + image_url