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

Подписаться 1,3 тыс.

Просмотров 23 тыс.

0% 0

In this comprehensive tutorial, Rama Castro, the Founder and CEO of Theos AI, walks you through the process of training the state-of-the-art YOLO V7 object detection model on Theos AI to detect and read license plates from vehicle images using Optical Character Recognition (OCR). Follow along to set up your account on Theos, upload and label your dataset, initiate the training, and deploy and test your model for real-world license plate recognition.
🚀 Dive right into this hands-on guide and get your License Plate Recognition system up and running with Theos AI!
#TheosAI #YOLOv7 #LicensePlateRecognition #ObjectDetection #OpticalCharacterRecognition #AI #MachineLearning #Tutorial
Chapters:
- 00:06 Introduction and Overview
- 00:24 Sign up on Theos AI
- 00:47 Exploring Object Detection Models on Theos AI
- 01:14 Creating and Uploading Dataset
- 02:01 Understanding Neural Network Training
- 05:49 Preparing Your Labeled Dataset
- 06:11 Initiating Training Session
- 09:27 Monitoring Training Metrics
- 09:51 Deploying and Testing Your Model
- 11:16 Performing OCR with Text Recognition Model
- 14:23 Conclusion and Resources
Links and Resources:
🥳 Sign up now for free: [Theos AI Signup](theos.ai/ref/y...)
👋🏻 Join our discord server: [Discord]( / discord )
✅ Join our WhatsApp group: [WhatsApp](chat.whatsapp....)
👨🏻‍💻 License Plate Recognition System source code: [GitHub](github.com/the...)
🤓 Read the blog post: [Blog](blog.theos.ai/...)
🚘 Download the license plates dataset: [Dataset](drive.google.c...)
📚 Read our official docs to learn how to use the Theos API in your software: [Docs](docs.theos.ai/...)
⚡️ Clone the Easy YOLOv7 repo to run it locally on any device without internet access in real-time performance: [Easy YOLOv7 GitHub Repo](github.com/the...)
👨🏻‍💼 Connect with Theos AI Team:
- Founder and CEO: [Rama Castro]( / ramacastro )
- Founder and President: [Nicolas Cantoli]( / nicolascantoli )
- COO: [Nawellopez95]( / nawellopez95 )
Don't forget to subscribe, like, and turn on notifications to stay updated with our latest tutorials. Leave a comment below on what tutorial you'd like to see next!

Опубликовано:

7 сен 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 59

@Melomano051984 5 месяцев назад

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

@shareb1t 9 дней назад

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

@amt.7rambo670 Год назад

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 Год назад

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

@nathakritnukulrat3063 Год назад

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

@theos-ai Год назад

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

@nathakritnukulrat3063 Год назад

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

@patawee313 Год назад

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

@theos-ai Год назад

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

@vkrts9176 Год назад

Awesome information brother

@theos-ai Год назад

Thank you! 🙌🏻

@kishoreg8835 Год назад

Excellent 🔥

@cyberhard Год назад

Great tutorial!

@theos-ai Год назад

Thanks!

@cyberhard Год назад

@@theos-ai you're welcome

@sanjaybhatikar 10 месяцев назад

I wasn unable to annotate bounding boxes for OCR as shown in the video <a href="#" class="seekto" data-time="259">4:19</a> - <a href="#" class="seekto" data-time="269">4:29</a>. 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 10 месяцев назад

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

@haideRMe Год назад

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

@theos-ai Год назад

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

@sanjaybhatikar 10 месяцев назад

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

@theos-ai 10 месяцев назад

Yes, you can do it in any language. Send us an email to contact@theos.ai explaining your use case, we will help you.

@RapidView Год назад

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

@theos-ai Год назад

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

@tsaashik8048 Год назад

in collab it asked <a href="#" class="seekto" data-time="491">8:11</a> 1. OBJECT DETECTION 2. OCR but u selected object detection.. we are doing ocr right then why selected object detection?

@theos-ai Год назад

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.

@amt.7rambo670 Год назад

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

@theos-ai Год назад

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

@vkrts9176 Год назад

Please make an tutorial on Handwritten text recognition.

@theos-ai Год назад

Sure, I will do it. Coming soon!

@user-gf3xy2fb7u Год назад

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 Год назад

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.

@lockieuxuan6454 5 месяцев назад

How can i use ocr for vietnamese

@eleniwang3414 Год назад

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 Год назад

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

@uchihamadara9363 11 месяцев назад

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 11 месяцев назад

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 11 месяцев назад

@@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 11 месяцев назад

@@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.

@cmdrelectrostruck1928 Год назад

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

@kishoreg8835 Год назад

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

@amt.7rambo670 Год назад

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 Год назад

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 Год назад

Send me a whatsapp message so I can better help you there: +41784208194

@amt.7rambo670 Год назад

@@theos-ai thanks bro waiting eagerly.

@mustafaates2096 Год назад

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 Год назад

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

@theos-ai Год назад

It's fixed now.

@amt.7rambo670 Год назад

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 Год назад

Hi, what do you mean by compulsory? We already have a video about integrating the deployment url to reactjs: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-n8ShYAtgqTk.html Please contact me on whatsapp so we can help you: +41784208194

@user-qm7ll8lg2m 9 месяцев назад

Its asking monthly and yearly plan for enrolling? 42$

@theos-ai 9 месяцев назад

Yes

@hawkingradiation3774 Год назад

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

@theos-ai Год назад

What problems did you run into?

@hawkingradiation3774 Год назад

@@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