hugging-apps/kraken-ppocrv6-medium-ocr

kraken PP-OCRv6 · medium

A Gradio demo for `small-models-for-glam/kraken-ppocrv6-medium`, a 15.9M-parameter PP-OCRv6 CTC text-line recogniser for kraken, trained on historical, contemporary and born-digital documents in 44 languages across 10 scripts (Arabic, Armenian, Cyrillic, Ethiopic, Georgian, Greek, Hebrew, Latin, Malayalam, Syriac).

The demo reproduces the model card's pipeline:

kraken -i image.png output.txt segment -bl ocr -m medium.safetensors

1. 1.Line detection — kraken's bundled BLLA baseline segmenter (segment -bl).
2. 2.Recognition — PP-OCRv6 medium runs over every detected line (ocr -m …).

Outputs are the line-detection overlay, the plain-text transcription, and an ALTO XML file with line and word coordinates.

Runs on ZeroGPU.

Example images

The three bundled examples are test fixtures from the kraken repository (Apache-2.0): a printed 19th-century French page, a Spanish handwritten notarial manuscript, and a printed Arabic page.

License

Model: Apache-2.0. kraken: Apache-2.0.