Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Whole line (multiple words) recognition

See original GitHub issue

Refer this issue https://github.com/emedvedev/attention-ocr/issues/59, are some modifications I make below sufficient to recognize multiple words (as a line)?

Include the space character into CHARMAP in aocr/util/data_gen.py.
Increase max-prediction for predicted phrase.
Increase max-width for input image

Issue Analytics

State:
Created 5 years ago
Comments:11 (5 by maintainers)

Top GitHub Comments

1reaction

ckirmsecommented, Apr 19, 2018

Hi @hiepph --sorry I didn’t see this message sooner. I’m using max-width 380, max-height 32, max-prediction 75 with a few tens of thousands of images. You don’t want max-height to be so large because all the images will get stretched to that size, and it will destroy the aspect ratio. You should be able to see some results with only 1000 images for training, but you need a lot more to get really good results I believe.

1reaction

ckirmsecommented, Apr 11, 2018

Yes that’s why --full-ascii exists too, @hiepph . I’ve done this successfully.

Top Results From Across the Web

python - Optical Character Recognition Multiple Line Detection

You want to recognize text of a document containing multiple lines. There are two ways to achieve this: Segment the document into lines...

The Science of Word Recognition - Typography | Microsoft Learn

This model says that the letters within a word are recognized simultaneously, and the letter information is used to recognize the words.

Phonics vs. Whole Word Reading - New Chapter Learning

For some children, phonics comes easily and they move into whole word recognition smoothly. For others, putting these sounds together is very difficult....

3. Word Recognition Skills: One of Two Essential Components ...

Our speech consists of whole words, but we write those words by breaking them down into their phonemes and representing each phoneme with...

What is OCR (Optical Character Recognition)? - AWS

Intelligent word recognition systems work on the same principles as ICR, but process whole word images instead of preprocessing the images into characters....