From Docker hell to Lite + LLM correction. A retrospective on my own experimentation, plus an introduction to someone else's browser-based NDLOCR-Lite implementation.
From browser OCR and server-side OCR to cloud APIs and AI — a roundup of what I learned trying to implement Japanese OCR on the web, including the limits of each approach.
When Layout Parser wouldn't install and NDLOCR alone couldn't handle a 4-column vertical text book, I used PyMuPDF and histogram analysis to brute-force split the columns.
