OCRopus 4 Inferece

This package implements OCRopus 4 inference.

Right now, there is a small command line program, a library, and a Jupyter Notebook demo.

• ocropus4 -- command line program for recognizing pages and lines
• ocropus4inf -- library implementing page recognition
• pagerec.py -- demo notebook

Comments:

• training and inference are two separate packages; training will be released separately
• the models are completely unoptimized, both in terms of runtime and in terms of error rates
• this uses loaded JIT models, which are slow the first few times around

Next Steps:

• optimize the models (memory, speed, error rate)
• transformer-based text line recognizer
• linearizer and/or bounding box extractor (using transformers)
• integrate deep-learning document cleanup