How to Train Paddle OCR on a Custom Dataset | YOLOv7 + OCR on Google Colab FREE GPU (ANPR/ALPR) 2023

[@Melomano051984]

U must have 1 million subscribers. Excellent. Alot of best

[@amt.7rambo670]

bro theos ai is gonna blow up. waiting eagerly for voice cloning. theosai makes everything easy with the interface. And bro are u the founder of theos ai?

[@theos-ai]

Thanks! We will upload new videos when voice cloning is ready. And yes, I'm the founder and CEO of Theos AI.

[@shareb1t]

Interesting, i wonder like can we teach this to Egyptology language for OCR into text?

[@nathakritnukulrat3063]

Can you help me when I train OCR why machine always disconnected? i have this problem many times.

[@theos-ai]

This is the way colab works, you should upgrade to Theos University to use our cloud GPUs that never disconnect.

[@nathakritnukulrat3063]

@@theos-ai colab gpu can't do this session?

[@kishoreg8835]

Excellent 🔥

[@vkrts9176]

Awesome information brother

[@theos-ai]

Thank you! 🙌🏻

[@haideRMe]

An error occurred Insufficient dataset examples. Please make sure you have labeled examples in the train, validation and test sets before training.

[@theos-ai]

You must label all your images before training. You must have labeled images in all data sets.

[@sanjaybhatikar]

Does it do Japanese Kanji? I have a use-case for pharmaceutical prescription reader.

[@theos-ai]

Yes, you can do it in any language. Send us an email to [e-mail] explaining your use case, we will help you.

[@cyberhard]

Great tutorial!

[@theos-ai]

Thanks!

[@cyberhard]

@@theos-ai you're welcome

[@patawee313]

Is it possible to classify to separate text type from the object? It might be great for doing key information extraction.

[@theos-ai]

Yes of course, you can make lots of text localization classes and then perform OCR on all of them.

[@eleniwang3414]

Hi I tried to run the code but got error 'Unable to log in with provided credentials. 'after login into my account, but I double checked it is indeed my password.

[@theos-ai]

Hi, were you able to solve it? you can try changing your password.

[@uchihamadara9363]

I would like to ask and this is based on my research in thesis creating system. Am I cheating if I use Theos AI because I am using a 3rd party access? Or will my research be invalidated?

[@theos-ai]

There's no problem if you use Theos AI for your thesis, many of our users are students doing their final year projects with our platform.

[@uchihamadara9363]

@@theos-ai please correct me, Theos AI will be only used to train model right? But the algorithm will still be Paddle OCR. Can you please describe what kind of algorithm will be used in Paddle OCR? I just crucially need it to be defined in my thesis. And can I get your permission that I will use Theos AI to be included in my documentation? This paper will be published.

[@theos-ai]

@@uchihamadara9363 Yes, Theos AI uses the official Paddle OCR model, here you have the paper: arxiv.org/pdf/2206.03001.pdf And yes, you have our permission to use Theos AI in your documentation.

[@mimich-q3c]

helloo i keep getting this error ERROR: CUDA out of memory. Tried to allocate 50.00 MiB (GPU 0; 14.75 GiB total capacity; 13.61 GiB already allocated; 44.81 MiB free; 13.74 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF can you help me with it ?

[@theos-ai]

Hello! have you tried reducing the batch size? The recommended batch size for YOLOv7 tiny in google colab T4 GPU is 32. Batch sizes should be in powers of 2: 128, 64, 32, 16, 8, 4, 2.

[@sanjaybhatikar]

I wasn unable to annotate bounding boxes for OCR as shown in the video 4:19 - 4:29. I looked in docs also and they show how to add bounding boxes for object detection but I did not see a reference to annotation of each bounding box. Unless I am missing something, is this a feature for a future release? Here is what I did: I went to datasets, created a new dataset, uploaded sample files, added a class 'text'. I was able to draw bounding boxes around text in each uploaded picture. But when I tried to annotate each bounding box, I did not see a way to do that. There is a grayed-out box that says "Confirm Skeleton" where I would expect a text field for annotation. Perhaps I am missing something. Thanks, - S.

[@theos-ai]

That was a bug, try reloading the website to get the updated code. It should work now.

[@cmdrelectrostruck1928]

What should I be changing if I try and do this with a very small dataset?

[@kishoreg8835]

Very small dataset won't work. You must use atleast 80k for ur desired result. This is minimum.

[@tsaashik8048]

in collab it asked 8:11 1. OBJECT DETECTION 2. OCR but u selected object detection.. we are doing ocr right then why selected object detection?

[@theos-ai]

If you watch the entire video you will understand that first we need to train the YOLOv7 tiny object detection model to locate the text within the image, and then we train PaddleOCR to read that text.

[@Tnetsinfineon]

Its asking monthly and yearly plan for enrolling? 42$

[@theos-ai]

Yes

[@lockieuxuan6454]

How can i use ocr for vietnamese

[@mustafaates2096]

hello my friend, i have a homework that i have to finish tomorrow and when i connect from theos to colab, it gives a lot of errors and doesn't do the train, can you try to fix it if i don't ask you please

[@kishoreg8835]

All issues fixed for yolo. OCR will be fixed in 2-3 days time

[@theos-ai]

It's fixed now.

[@amt.7rambo670]

bro that ocr package is available in npm ? and in docs we need code for react js ocr integration also mate.

[@theos-ai]

Won't be needed in npm. For reactjs you will be able to deploy both YOLOv7 and custom paddle ocr models on the same deployment URL. Then you can use that URL like this: docs.theos.ai/deployment#react

[@amt.7rambo670]

Bro as u said i did first trained yolov7 tiny for license plate detection then trained another training session for paddle ocr same dataset. Now i deploy it on yolov7 algorithm. But in playground i am not seeing ocr text detction?

[@theos-ai]

Yes I know, on the playground currently you can select one of the default OCR models. We'll soon add the ability to deploy your own OCR model to use on the playground and API. For now you can use it as I show in this video, with the Easy YOLOv7 and Easy Paddle OCR GitHub repositories. github.com/theos-ai/easy-paddle-ocr

[@theos-ai]

Send me a whatsapp message so I can better help you there: [phone]

[@amt.7rambo670]

@@theos-ai thanks bro waiting eagerly.

[@RapidView]

Hi Dear Tnx for amazing tutorials. How can we do that for tessaract

[@theos-ai]

Thanks! check this out: saiashish90.medium.com/training-tesseract-ocr-with-custom-data-d3f4881575c0

[@amt.7rambo670]

bro is fallback url compulsory and pls my friend can u make a video on how to integrate deployment url to react js and its working. I have paid version of theos ai but no one response. so icome to u. i have project to submit for hackathon within a week.

[@theos-ai]

Hi, what do you mean by compulsory? We already have a video about integrating the deployment url to reactjs: kzbin.info/www/bejne/pGm2mYx3qcykis0 Please contact me on whatsapp so we can help you: [phone]

[@vkrts9176]

Please make an tutorial on Handwritten text recognition.

[@theos-ai]

Sure, I will do it. Coming soon!

[@tsaashik8048]

is this free to use?

[@theos-ai]

Yes, we have a forever free plan, it comes with 1000 image capacity, unlimited trainings and 1 deployment.

[@merasaleh3190]

Can the plates have non English alphabet Like Arabic??

[@theos-ai]

Hi 👋🏻 yes of course!

[@merasaleh3190]

Thanks for responding

[@hawkingradiation3774]

has anyone been successfully able to install paddleocr on m1 mac? if yes then pls tell how

[@theos-ai]

What problems did you run into?

[@hawkingradiation3774]

@@theos-ai for some reason pymupdf wheel doesnt get created when installing paddleocr, paddlepaddle gets installed successfully but installing paddleocr using the command provided in the github repo link doesnt work, i create new env for each project and used python3.9 as well as python3.10 for this