⚠️Warning

• Transformers versions above 4.28.XX may break inference result: https://github.com/huggingface/transformers/issues/39473
• Latest versions seem to work properly (e.g 4.57.6)

Donut German Receipts

Model Details

• Base model: naver-clova-ix/donut-base
• Dataset:
  • german receipts based on small private dataset(473 images)
• Training:
  • training based on 435 images
  • test validation with 38 images
    • accurcay on supermarked-receipts is well (~70% exact string comparison)
    • not so many unkown stores in dataset, results differ in quality
  • Image size used: 960x1280
  • Epochs: 3
  • LR: 2e-5
  • Batch-Size: 2

Input Example

Output Schema

Framework versions

• Transformers 4.27.4
• Pytorch 2.7.1
• Datasets 4.0.0
• Tokenizers 0.13.3