把扫描版图书 先变成可读的 EPUB 再决定要不要继续做 整本转换。

The live checker runs the repo's quality rules against extracted page text. It flags empty OCR output, page-number leaks, broken hyphenation, bad spacing, and formula-structure risk.

The current demo can generate a preview EPUB directly from an uploaded PDF. Production-grade full-book conversion still needs stronger layout recovery, better OCR repair, EPUB validation, and job orchestration.

In this demo, the PDF is uploaded to the current preview service so it can generate a sample EPUB. Review your deployment and privacy settings before testing private material. The page checker still exists for users who want to inspect extracted text before uploading.

The wedge is scanned PDF input, extracted page text for diagnosis, and a reflowable EPUB preview for reading on Kindle- and Kobo-style devices. It is not trying to become a broad everything-to-everything converter.

Accurate enough to judge reading comfort and obvious structural damage. The preview is meant to answer whether the book feels clean enough to continue, not to replace final editorial review for every edge case.

Some cleanup can be automated, but the product should stay honest about uncertainty. The key promise is to surface where review is needed so the user can decide whether to repair, crop, rerun OCR, or stop.

Those are the pages most likely to trigger risk labels. Formula structure, tabular alignment, footnotes, and badly cropped scans often need targeted repair before an EPUB is truly comfortable to read.

Readers and document owners with scanned books, academic PDFs, or public-domain material who want a reflowable EPUB for Kindle or Kobo without proofreading every page manually.

Generic converters export files, but they rarely explain where OCR or layout recovery failed. This product is designed to show risk before the user commits to a full conversion.

Yes. The main demo flow is designed around that exact question: upload one PDF, inspect the first pages, review OCR risk, and judge whether the full book is worth converting.

The most useful examples are before-and-after comparisons for noisy OCR, math-heavy academic pages, two-column journal layouts, footnotes, captions, and image-only scans that need OCR fallback.

It should prove reading comfort, not just file export. A good preview shows whether line breaks, page numbers, formulas, headings, columns, and captions still make sense on a small reading device.