donut-ktp-extractor

Donut KTP OCR Model

This model is fine-tuned from Donut for extracting information from Indonesian ID Cards (KTP).

Model Description

• Model type: Donut (Document Understanding Transformer)
• Language: Indonesian
• Task: OCR and Information Extraction from KTP
• Accuracy: TED: 90.97%, F1: 84.28%
• Fields extracted: 15 fields

Extracted Fields

The model can extract the following fields from KTP:

1. NIK (ID Number)
2. Nama (Name)
3. Tempat Lahir (Place of Birth)
4. Tanggal Lahir (Date of Birth)
5. Jenis Kelamin (Gender)
6. Golongan Darah (Blood Type)
7. Alamat (Address)
8. RT/RW
9. Kelurahan (Village)
10. Kecamatan (District)
11. Agama (Religion)
12. Status Perkawinan (Marital Status)
13. Pekerjaan (Occupation)
14. Kewarganegaraan (Citizenship)
15. Berlaku Hingga (Valid Until)

Usage

import torch
from PIL import Image
from donut import DonutModel

# Load model
model = DonutModel.from_pretrained("ahmadarif019/donut-ktp-extractor")
model.eval()

# Use GPU if available
if torch.cuda.is_available():
    model.half()
    model.to("cuda")

# Process image
image = Image.open("ktp.jpg").convert("RGB")
result = model.inference(image=image, prompt="<s_dataset_ktp>")
ktp_data = result["predictions"][0]

print(ktp_data)

Training Data

• Dataset: Indonesian KTP images
• Size: [Add your dataset size]
• Split: [Add train/val/test split info]

Performance

• TED (Tree Edit Distance): 90.97%
• F1 Score: 84.28%

Limitations

• Works best with clear, well-lit KTP images
• May have reduced accuracy with damaged or obscured cards
• Optimized for standard KTP format

Citation

@article{kim2021donut,
  title={OCR-free Document Understanding Transformer},
  author={Kim, Geewook and Hong, Teakgyu and Yim, Moonbin and Nam, JeongYeon and Park, Jinyoung and Yim, Jinyeong and Hwang, Wonseok and Yun, Sangdoo and Han, Dongyoon and Park, Seunghyun},
  journal={arXiv preprint arXiv:2111.15664},
  year={2021}
}

License

MIT License - For research and internal use only.