Available from version 0.1.11. The Windows version is updated by the Microsoft Store. On a Mac, download the new version from Releases.
Watch
A page of a classical Chinese book read with PaddleOCR-VL and PP-OCRv6, with the highlighted characters checked against the page image (about 1 minute, with narration and captions).
What changed
Press “Compare” at the top and tick two or more engines. Characters that the engines read differently get a yellow background and an underline. Where nothing is highlighted, every engine read the same.
- In the “In use” column, a character is highlighted if any other engine read it differently
- In the other columns, a character is highlighted if it differs from the “In use” result
- Each column shows how much it agrees (for example “96%”) and how many characters differ
- To turn the highlighting off, switch off “Highlight differences” at the top
Check highlighted characters against the page
A highlight only means the readings differ. It doesn’t say which one is right. On the page in the video, the highlights fell into three kinds:
- Text outside the main text: one engine also read the ruler and the library label in the photo
- Misreadings: for example 機栝 read as 機恬. The page image shows which is right
- Variant forms: 別 and 别 are the same character in different forms. Not necessarily an error
Also changed
The label on the result used for copying and saving is now “In use”. For a result without line positions (no boxes on the image) it reads “In use (no positions)”.
Other changes in this version are in the changelog.