183 - OCR in python using keras-ocr

[@IntenseRouge]

Wow! I wasn’t aware of this ocr library and not having been satisfied with the results of what I had tried out, I am so glad you posted this video. I’ll test the keras ocr soon. Thank you so much.

[@DigitalSreeni]

Welcome!

[@deepakkumarjha1442]

Really amazing work sir. I was putting huge effort for building the OCR like this in very short span of time and while working on it I spotted this and the work is done. Now I move further in other parts of the research. Great!!! Thanks again sir...

[@DigitalSreeni]

I am glad it helped. The whole point of my tutorials is to teach but also help others not waste their time reinventing code.

[@akshaykhare9466]

I tried this code, but after printing np.shape(images), every time it is showing message 'kernel died' . This is happening in Jupyter notebook as well as in Spyder. So I am not getting the output. I tried restarting kernel but same thing happening again and again. What should I do?

[@arturosorianorodriguez4584]

I have a problem. The code recognizes all the letters, but the H no. What can I do?

[@shimisys429]

when i write the code "import keras_ocr", it giving me an error said that no module named 'keras_ocr'. I already pip install the keras-ocr

[@shimisys429]

try: # %tensorflow_version only exists in Colab. %tensorflow_version 2.x except Exception: pass I just found the solution. need to install tensorflow first with this code

[@sagihaviv5675]

sir is it possible to recognize and extract a specific sentence or word in ocr? do i need to use nlp for that? please give me advice

[@truongzgiang]

Only trian for show , can't do anything useful with it , if you want to take words form image , you write it with fingers , believe me 1000 ocr video on KZbin is trash , only trian and do nothing

[@kusamuzukashi3825]

amazing video sir, thanks for this nice explaining

[@shreyaputhran7646]

how to print the text as output. Pls help

[@sharonibejih2806]

Hi @DigitalSreeni, I'm trying to finetune the keras OCR recogniser using my own dataset by following their documentation. But my image size is different from theirs and this is giving an error (as anticipated). The error message says ValueError: You are trying to load a weight file containing 16 layers into a model with 15 layers. I am just finding it hard to figure out how to change the input size, without having to train from scratch. Can you help, please? Looking forward to your response.

[@DigitalSreeni]

Please checkout if this approach works for you. kzbin.info/www/bejne/pajMopewncaSeqc

[@ouardafenek3765]

Dear Sir, Thank you for the video. Is it possible to fine tune the model on a personal data to make it better on use case ? Thank's in advance.

[@DigitalSreeni]

I recommend you to look into the documentation for this library to fine tuning the model. Often times you should be able to take their model and further train it with your data.

[@muhammadizzat4388]

sorry to ask, how do i install it from github??

[@DigitalSreeni]

Please read the documentation on github. It saus, one way to install is via: pip install keras-ocr

[@dileepchakravarthyhsa6351]

hi sreeni i am trying to run code but i am getting this error again and agian can you help me to sort it out TypeError: annotate() missing 1 required positional argument: 's'

[@DigitalSreeni]

This seems to be a new error showing up when you try to draw extracted text on to your original image. But the actual process of extracting text should work.

[@Zeeshanmys]

Great information srini.. Which has better accuracy is it keras ocr or tessaract...??

[@DigitalSreeni]

I like EasyOCR... wait for my video on this topic... (August 2nd)

[@Zeeshanmys]

@@DigitalSreeni that was quick.. Thanks srini... Waiting for it..

[@roderickdunn2517]

Thank you for the great introduction to this library, it looks fantastic. However, running the demo code on my Mac (M1) is giving me strange results - Text Detection and Localization is working fine, but the character recognitions are completely wrong, most words are recognized as something like "tuueldkstt" or other very similar nonsense. Almost seems like the recognition model wasn't loaded correctly. Do you have any thoughts as to why this could be happening?

[@evyatarcoco]

Dear sir, Where can we subscribe to every keras library released in order to keep learning? Do you have any recommendations? Thanks

[@DigitalSreeni]

Unfortunately there is not single place (that I know of) where you can get information about all keras based applications. May be keras documentation page helps a bit but many useful applications based on kears are not listed there.

[@mohamedtouati9663]

Is keras-ocr work with other languages? I just see no argument in the function

[@DigitalSreeni]

I believe it is English only. Try tesseract for non English OCR. I never used it but the documentation seems to be straightforward.

[@mohamedramzyibrahim6375]

Thank you very much

[@DigitalSreeni]

Any time

[@moh92001]

thanks

[@DigitalSreeni]

Welcome

[@buxA57]

After line pipeline.recognize(images) no more code is executed and program finishes, i have no idea what is problem even after google search. 2022-06-23 12:24:26.055634: I tensorflow/stream_executor/cuda/cuda_dnn.cc:384] Loaded cuDNN version 8400 Process finished with exit code -1073740791 (0xC0000409)