I will update/edit this issue to track development process of new language. The current list is

Group 1 (Arabic script)

• Arabic (DONE, August, 5 2020)
• Uyghur (DONE, August, 5 2020)
• Persian (DONE, August, 5 2020)
• Urdu (DONE, August, 5 2020)

Group 2 (Latin script)

• Serbian-latin (DONE, July,12 2020)
• Occitan (DONE, July,12 2020)

Group 3 (Devanagari)

• Hindi (DONE, July,24 2020)
• Marathi (DONE, July,24 2020)
• Nepali (DONE, July,24 2020)
• Rajasthani (NEED HELP)
• Awadhi, Haryanvi, Sanskrit (if possible)

Group 4 (Cyrillic script)

• Russian (DONE, July,29 2020)
• Serbian-cyrillic (DONE, July,29 2020)
• Bulgarian (DONE, July,29 2020)
• Ukranian (DONE, July,29 2020)
• Mongolian (DONE, July,29 2020)
• Belarusian (DONE, July,29 2020)
• Tajik (DONE, April,20 2021)
• Kyrgyz (NEED HELP)

Group 5

• Telugu (DONE, November,17 2020)
• Kannada (DONE, November,17 2020)

Group 6 (Language that doesn't share characters with others)

• Tamil (DONE, August, 10 2020)
• Hebrew (ready to train)
• Malayalam (ready to train)
• Bengali + Assamese (DONE, August, 23 2020)
• Punjabi (ready to train)
• Abkhaz (ready to train)

Group 7 (Improvement and possible extra models)

• Japanese version 2 (DONE, March, 21 2021)+ vertical text
• Chinese version2 (DONE, March, 21 2021)+ vertical text
• Korean version 2(DONE, March, 21 2021)
• Latin version 2 (DONE, March, 21 2021)
• Math + Greek?
• Number+symbol only

Guideline for new language request

To request a new language support, I need you to send a PR with 2 following files

1. In folder easyocr/character, we need 'yourlanguagecode_char.txt' that contains list of all characters. Please see format/example from other files in that folder.
2. In folder easyocr/dict, we need 'yourlanguagecode.txt' that contains list of words in your language. On average we have ~30000 words per language with more than 50000 words for popular one. More is better in this file.

If your language has unique elements (such as 1. Arabic: characters change form when attach to each other + write from right to left 2. Thai: Some characters need to be above the line and some below), please educate me with your best ability and/or give useful links. It is important to take care of the detail to achieve a system that really works.

Lastly, please understand that my priority will have to go to popular language or set of languages that share most of characters together (also tell me if your language share a lot of characters with other). It takes me at least a week to work for new model. You may have to wait a while for new model to be released.

Hello! Thanks for that amazing library first of all! Could someone please help to resolve the issue i encountered today only (yesterday and before it was working smoothly).

in my code i have let's say:

import easyocr
reader = easyocr.Reader(['id', 'en'])

When i run it - i am getting the following error:

CUDA not available - defaulting to CPU. Note: This module is much faster with a GPU.
MD5 hash mismatch, possible file corruption
Re-downloading the recognition model, please wait
Traceback (most recent call last):
  File "tryout_easyocr.py", line 5, in <module>
    reader = easyocr.Reader(['id', 'en'])
  File "/usr/local/lib/python3.7/site-packages/easyocr/easyocr.py", line 194, in __init__
    urlretrieve(model_url[model_file][0], MODEL_PATH)
  File "/usr/local/Cellar/python/3.7.8/Frameworks/Python.framework/Versions/3.7/lib/python3.7/urllib/request.py", line 288, in urlretrieve
    % (read, size), result)

Regardless what language i choose - i face this error in all the environments:

• in mac os runtime
• in docker
• in ubuntu
• in colab https://colab.research.google.com/github/vistec-AI/colab/blob/master/easyocr.ipynb#scrollTo=lIYdn1woOS1n

Diving deeper it tries to download the following file: https://www.jaided.ai/read_download/latin.pth which i wasn't able to download with wget, curl or browser as well for the same issue.

Seems https://www.jaided.ai/ resets the connection during download

I am facing issue with easyocr.reader module. I have successfully imported easyocr, but face issue on following line.

reader = easyocr.Reader(['ch_sim', 'en'])

error is following.

AttributeError: module 'easyocr' has no attribute 'Reader'

Looks like there's an issue with the server for the model files. Returns a 403 forbidden when attempting to download.

    reader = easyocr.Reader(['en'], gpu = False)
  File "/Users/travis/.local/share/virtualenvs/ocr-RN8nrvRp/lib/python3.8/site-packages/easyocr/easyocr.py", line 185, in __init__
    urlretrieve(model_url['detector'][0] , DETECTOR_PATH)
  File "/usr/local/opt/[email protected]/Frameworks/Python.framework/Versions/3.8/lib/python3.8/urllib/request.py", line 247, in urlretrieve
    with contextlib.closing(urlopen(url, data)) as fp:
  File "/usr/local/opt/[email protected]/Frameworks/Python.framework/Versions/3.8/lib/python3.8/urllib/request.py", line 222, in urlopen
    return opener.open(url, data, timeout)
  File "/usr/local/opt/[email protected]/Frameworks/Python.framework/Versions/3.8/lib/python3.8/urllib/request.py", line 531, in open
    response = meth(req, response)
  File "/usr/local/opt/[email protected]/Frameworks/Python.framework/Versions/3.8/lib/python3.8/urllib/request.py", line 640, in http_response
    response = self.parent.error(
  File "/usr/local/opt/[email protected]/Frameworks/Python.framework/Versions/3.8/lib/python3.8/urllib/request.py", line 569, in error
    return self._call_chain(*args)
  File "/usr/local/opt/[email protected]/Frameworks/Python.framework/Versions/3.8/lib/python3.8/urllib/request.py", line 502, in _call_chain
    result = func(*args)
  File "/usr/local/opt/[email protected]/Frameworks/Python.framework/Versions/3.8/lib/python3.8/urllib/request.py", line 649, in http_error_default
    raise HTTPError(req.full_url, code, msg, hdrs, fp)
urllib.error.HTTPError: HTTP Error 403: Forbidden

Hi,

Is the current version support digit recognition? If not, please add in future release. The OCR to recognize digits from meter is a common but will be very useful case. Sample cropped image of a meter screen-

Currently it is giving following result for above sample image-

Hi, I am working on a project where I need the bounding boxes around the characters rather than the whole word. Is there any way I can do that using EasyOCR? Thanks

Hi! I'm a student who work in common computer as intern. recently I'm working for deploying any useful and interesting project by using Ainize. and I ainize easyocr project so that every user can experience easyocr easy! Try click ainize button or visit webpage! you can try it by click ainize button or [web page] link in README.md thank you!

(Ainize is a project that can deploy server for free & able to access live services from Github with one click! if you interested in Ainize, try visit ainize.ai !)

Have a good day 😄!

This project is impressive. I've tested this and it's extremely more precise than Tesseract, thank you for your effort.

There are a few issues I've noticed on some test cases I've tested, some I think are caused by missing symbols, others are more specific to the Japanese language, and could be improved from the context of a dictionary (looks like ja has only characters right now).

Is there anyway I can help you? I can certainly add the missing characters to the characters list, and I'm willing to also build a dictionary if that could help disambiguating some words. But I would have to wait for you to retrain the model on your side?

Here are my test cases:

1 - いや…あるというか…って→

https://0x0.st/ivNP.png

issues:

Missing … missing → mistakes って for つて (fixable with a dictionary file I think)

result:

([[208, 0], [628, 0], [628, 70], [208, 70]], 'あるというか:', 0.07428263872861862)
([[0, 1], [185, 1], [185, 69], [0, 69]], 'いや・', 0.2885110080242157)
([[3, 69], [183, 69], [183, 128], [3, 128]], 'つて', 0.4845466613769531)

2 - ♬〜（これが私の生きる道）

https://0x0.st/ivZC.png

issues:

Missing ♬〜 Mistakes （これ for にれ Mistakes が(ga) for か(ka) Detects English parens () instead of Japanese parens （）

([[1, 0], [125, 0], [125, 63], [1, 63]], ',~', 0.10811009258031845)
([[179, 0], [787, 0], [787, 66], [179, 66]], 'にれか私の生きる道)', 0.3134567439556122)

3 - （秀一）ああッ…もう⁉︎

https://0x0.st/ivZh.png

issues:

Mistakes small ッ for big ツ (similar to 1 but katakana instead of hiragana) Mistakes … for ・・・

([[0, 0], [174, 0], [174, 64], [0, 64]], '(秀一)', 0.9035432934761047)
([[207, 0], [457, 0], [457, 64], [207, 64]], 'ああツ・・・', 0.35586389899253845)
([[481, 0], [668, 0], [668, 64], [481, 64]], 'もう!?', 0.4920879304409027)

4 - そっか

https://0x0.st/ivZ7.png

issues:

mistakes そっか for そつか (fixable with a dictionary file, I think) (similar to 1 そっか is a really common word)

([[0, 0], [186, 0], [186, 60], [0, 60]], 'そつか', 0.9190227389335632)

5 - （久美子）うん　ヘアピンのお礼

https://0x0.st/ivZR.png

issues:

mistakes の for 0 (not sure how to fix this one, but it's pretty important – seems like の is properly recognized in my test case 2) mistakes ヘアピン for へアピソ (fixable by dictionary probably)

([[0, 0], [238, 0], [238, 72], [0, 72]], '(久美子)', 0.9745591878890991)
([[268, 0], [396, 0], [396, 70], [268, 70]], 'うん', 0.5724520087242126)
([[22, 60], [454, 60], [454, 132], [22, 132]], 'へアピソ0お礼', 0.25971919298171997)

Thanks for publishing this great EASYOCR model! I am wondering if I can find a tutorial to train EASYOCR or finetune it on a custom dataset ( where I need to add a complex background for texts and support new fonts).

What do you think? is there any link for that?

Any update?

I've been trying to read some Sudokus and so far all digits have been recognized except 7. Is this a font issue maybe? All the images I have tried, no 7 was recognized.

Edit: Maybe related #130

Example:

Link to the image

 curl -o sudoku.png https://upload.wikimedia.org/wikipedia/commons/thumb/f/ff/Sudoku-by-L2G-20050714.svg/1200px-Sudoku-by-L2G-20050714.svg.png
 easyocr -l en -f sudoku.png

([[42, 34], [100, 34], [100, 114], [42, 114]], '5', 0.9198901653289795)
([[174, 34], [234, 34], [234, 114], [174, 114]], '3', 0.5189616680145264)
([[704, 166], [762, 166], [762, 246], [704, 246]], '5', 0.8919380307197571)
([[44, 168], [104, 168], [104, 246], [44, 246]], '6', 0.8406975865364075)
([[568, 168], [630, 168], [630, 246], [568, 246]], '9', 0.7890614867210388)
([[445, 175], [499, 175], [499, 243], [445, 243]], '_', 0.8146840333938599)
([[176, 300], [234, 300], [234, 376], [176, 376]], '9', 0.8461765646934509)
([[968, 302], [1026, 302], [1026, 378], [968, 378]], '6', 0.9359095692634583)
([[44, 432], [102, 432], [102, 510], [44, 510]], '8', 0.9884797930717468)
([[1098, 432], [1156, 432], [1156, 510], [1098, 510]], '3', 0.5310271382331848)
([[572, 434], [630, 434], [630, 510], [572, 510]], '6', 0.9405879974365234)
([[46, 564], [108, 564], [108, 638], [46, 638]], '4', 0.6061235070228577)
([[702, 564], [762, 564], [762, 642], [702, 642]], '3', 0.617751955986023)
([[1100, 566], [1158, 566], [1158, 640], [1100, 640]], '_', 0.9189348220825195)
([[441, 569], [493, 569], [493, 637], [441, 637]], '8', 0.4214801788330078)
([[570, 694], [628, 694], [628, 772], [570, 772]], '2', 0.9453338384628296)
([[1098, 696], [1158, 696], [1158, 772], [1098, 772]], '6', 0.8925315737724304)
([[834, 826], [894, 826], [894, 906], [834, 906]], '2', 0.9729335308074951)
([[176, 830], [236, 830], [236, 906], [176, 906]], '6', 0.9366582036018372)
([[966, 830], [1024, 830], [1024, 904], [966, 904]], '8', 0.9889897704124451)
([[700, 956], [762, 956], [762, 1036], [700, 1036]], '9', 0.7338930368423462)
([[1098, 958], [1156, 958], [1156, 1036], [1098, 1036]], '5', 0.9352748394012451)
([[439, 961], [505, 961], [505, 1029], [439, 1029]], '4', 0.5053627490997314)
([[572, 962], [634, 962], [634, 1038], [572, 1038]], '_', 0.8581225872039795)
([[570, 1090], [630, 1090], [630, 1168], [570, 1168]], '8', 0.9891173839569092)
([[1096, 1090], [1154, 1090], [1154, 1166], [1096, 1166]], '9', 0.6340051293373108)

hi , I have a question, when i execute pip install easyocr ,tips : a error: could not find a version that satisfies the requirement thorchvision>=0.5(from easyocr)

Hi. I am currently trying to train my own model. I obtained the dataset as required and modified the config file. However, I get this error when trying to train. I have already tried to decrease workers in config file. I also tried to modify the dataset.py file, line 101 from image, text = data_loader_iter.next() to image, text = next(data_loader_iter)

However, the error persist.

Thanks

Hello! I am a developer who is working on various projects using EasyOCR.

I am sending a PR to suggest a function that would be helpful to many people after doing several experiments.

During detection, data with incorrectly configured coordinate vector values is free_list, The horizontal_list has been confirmed to go to data consisting of exact coordinate vector values.

However, this comes as horizontal_list + free_list when receiving the result, so if you need to check the data sequentially, you need to compare the pixels directly and see where they are recognized.

The gif uploaded above is an image that shows that when recognized using EasyOCR directly, the data comes in sequentially and 2 and 4 come in last.

This means that the free_list is not sorted at the end and is merged as it is.

This is difficult to see at a glance even when detail=0 is inserted into the detail parameter.

So I developed a function that makes it easier to see by aligning the horizontal_list and free_list together when free_merge is inserted into the output_format parameter.

The gif uploaded above is an image that returns the result value sequentially after adding the function to EasyOCR.

result = reader.readtext(image, output_format='free_merge')
for r in result:
	print(r)

If you enter as above, you want it to be an EasyOCR that returns the results sequentially as follows.

([[90, 72], [162, 72], [162, 172], [90, 172]], '1', 0.9979992393730299)
([[299, 53], [387, 53], [387, 185], [299, 185]], '2', 1.0)
([[522, 44], [598, 44], [598, 172], [522, 172]], '3', 0.9944517558838641)
([[745, 53], [831, 53], [831, 169], [745, 169]], '4', 0.9790806838048891)
([[968, 52], [1042, 52], [1042, 174], [968, 174]], '5', 1.0)
([[1188, 54], [1266, 54], [1266, 172], [1188, 172]], '6', 0.9999949932161059)
([[1415, 77], [1475, 77], [1475, 169], [1415, 169]], '7', 0.9960819788421169)
([[1626, 54], [1706, 54], [1706, 174], [1626, 174]], '8', 1.0)
([[1848, 64], [1920, 64], [1920, 174], [1848, 174]], '9', 0.9999967813517721)
([[2027, 43], [2185, 43], [2185, 184], [2027, 184]], '10', 0.9999989884757856)
([[77.2879532910169, 230.02821146681296], [151.37898588609144, 239.7842872259749], [132.7120467089831, 373.971788533187], [58.62101411390856, 364.2157127740251]], '2', 0.9868415024810453)
([[281, 199], [391, 199], [391, 365], [281, 365]], '2', 0.9980995156509067)
([[526, 236], [574, 236], [574, 350], [526, 350]], '1', 0.3553702823128333)
([[738, 226], [836, 226], [836, 372], [738, 372]], '4', 1.0)
([[872, 282], [904, 282], [904, 358], [872, 358]], '1', 0.46445868490119224)
([[920.8651368309256, 237.1345809643125], [1041.2936593026684, 188.84521212806337], [1093.1348631690744, 349.86541903568747], [972.7063406973315, 398.15478787193666]], '4', 1.0)
([[1162, 224], [1266, 224], [1266, 384], [1162, 384]], '3', 0.9999594692522464)
([[1365, 213], [1497, 213], [1497, 407], [1365, 407]], '5', 0.9986469557185416)
([[1588, 200], [1714, 200], [1714, 394], [1588, 394]], '6', 0.3138604097965505)
([[1853, 255], [1893, 255], [1893, 365], [1853, 365]], '1', 0.9972939940858829)
([[2075, 239], [2117, 239], [2117, 339], [2075, 339]], '1', 0.9854363293399651)
([[47, 439], [199, 439], [199, 575], [47, 575]], '11', 0.9999839842351304)
([[264, 434], [422, 434], [422, 578], [264, 578]], '12', 0.9999954481433951)
([[489, 437], [639, 437], [639, 577], [489, 577]], '13', 0.9999845742882926)
([[709, 437], [865, 437], [865, 577], [709, 577]], '14', 0.9997981225645103)
([[929, 441], [1083, 441], [1083, 579], [929, 579]], '15', 0.9425667175676142)
([[1151, 445], [1303, 445], [1303, 579], [1151, 579]], '16', 0.9999962910793456)
([[1368, 445], [1516, 445], [1516, 579], [1368, 579]], '17', 0.999997049721879)
([[1589, 445], [1741, 445], [1741, 579], [1589, 579]], '18', 0.9999982298328214)
([[1809, 447], [1961, 447], [1961, 585], [1809, 585]], '19', 0.9999972183091315)
([[2031, 445], [2183, 445], [2183, 581], [2031, 581]], '20', 0.9999991570631338)
([[50, 630], [172, 630], [172, 794], [50, 794]], '4', 1.0)
([[260, 622], [428, 622], [428, 798], [260, 798]], '4', 0.9993658331357935)
([[494, 618], [598, 618], [598, 782], [494, 782]], '3', 0.9969145055207527)
([[719, 621], [831, 621], [831, 781], [719, 781]], '5', 0.999999880790714)
([[949, 623], [1041, 623], [1041, 773], [949, 773]], '2', 0.9640018726844447)
([[1173, 655], [1239, 655], [1239, 753], [1173, 753]], '1', 0.9843721660900542)
([[1405, 633], [1471, 633], [1471, 767], [1405, 767]], '1', 0.99952905955627)
([[1606, 628], [1704, 628], [1704, 784], [1606, 784]], '2', 0.9996682680632638)
([[2039, 623], [2151, 623], [2151, 801], [2039, 801]], '2', 0.31963881498015567)
([[43, 845], [196, 845], [196, 979], [43, 979]], '21', 0.9999989041821146)
([[264, 841], [416, 841], [416, 981], [264, 981]], '22', 0.9999998314126102)
([[487, 843], [635, 843], [635, 981], [487, 981]], '23', 0.9999978083645809)
([[707, 841], [863, 841], [863, 981], [707, 981]], '24', 0.9999994942378553)
([[928, 840], [1082, 840], [1082, 984], [928, 984]], '25', 0.9999996628252286)
([[1152, 848], [1300, 848], [1300, 976], [1152, 976]], '26', 0.9864728654305385)
([[1369, 843], [1517, 843], [1517, 981], [1369, 981]], '27', 0.6750208001814506)
([[1589, 847], [1741, 847], [1741, 983], [1589, 983]], '28', 0.9999988299663297)
([[1811, 849], [1961, 849], [1961, 987], [1811, 987]], '29', 0.9999996628252286)
([[2032, 852], [2180, 852], [2180, 980], [2032, 980]], '30', 0.9999972183091315)
([[47, 1021], [183, 1021], [183, 1193], [47, 1193]], '5', 0.9999997615814351)
([[275, 1033], [385, 1033], [385, 1191], [275, 1191]], '2', 0.9999992847443906)
([[488, 1028], [610, 1028], [610, 1198], [488, 1198]], '5', 0.999989390401371)
([[724, 1022], [820, 1022], [820, 1174], [724, 1174]], '3', 0.16538231022019545)
([[927, 1013], [1043, 1013], [1043, 1191], [927, 1191]], '3', 0.9998320411641579)
([[1812, 1030], [1986, 1030], [1986, 1180], [1812, 1180]], '4', 0.9999662640555904)
([[2025, 1031], [2163, 1031], [2163, 1173], [2025, 1173]], '4', 1.0)

Originally, I added this feature to use, but I'm sure it'll be a necessary feature for someone.

Thank you for reading the long article.

I tried to recognize number plates from images of cars. The OCR fails to detect when the number is tilted. What should be done to improve the detection accuracy?

I was using a custom dataset to train EasyOCR, which has seven segment display digits. The training worked well up until the 110000th iteration, however my runtime got disconnected in between and I now want to restart it from the saved checkpoint. So, could someone please explain or assist me ?