Recognition being run even when disabled #769

Balearica · 2023-05-29T18:21:14Z

Tesseract.js skips the recognition step when all of the output formats requested do not require recognition. This allows for using Tesseract.js for image processing without waiting for recognition (the step with the longest runtime) to be completed. This behavior will also be critical if we add a way to get layout data without running recognition (requested in #656).

tesseract.js/src/worker-script/index.js

Lines 355 to 359 in 15fdd9e

    
           if (recOutputCount > 0) { 
        
             api.Recognize(null); 
        
           } else { 
        
             log('Skipping recognition: all output options requiring recognition are disabled.'); 
        
           }

Unfortunately, this does not appear to be working correctly at present--recognition is still being run even when we intentionally skip running api.Recognize. It looks like something in the dump function is triggering Tesseract to automatically run recognition under the hood.

The text was updated successfully, but these errors were encountered:

Balearica · 2023-05-29T18:41:28Z

The cause ended up being the api.MeanTextConf function, which was always run by dump and appears to automatically trigger recognition.

tesseract.js/src/worker-script/utils/dump.js

Line 215 in b23ebf3

confidence: api.MeanTextConf(),

Resolved by #770. Will be included in next release.

Balearica mentioned this issue May 29, 2023

Add ability to analyse layout without running recognition #770

Merged

Balearica closed this as completed May 29, 2023

Recognition being run even when disabled #769

Recognition being run even when disabled #769

Balearica commented May 29, 2023

Balearica commented May 29, 2023 •

edited

Recognition being run even when disabled #769

Recognition being run even when disabled #769

Comments

Balearica commented May 29, 2023

Balearica commented May 29, 2023 • edited

Balearica commented May 29, 2023 •

edited