Help Needed: Setting Up a Virtual Environment on Ubuntu for OCR Workflows

AI Discussions
1

Good morning friends,

I need your guidance in setting up a virtual environment on my Linux system (Ubuntu 22.04 LTS – Jammy Jellyfish) that can handle OCR tasks efficiently without any package conflicts. Below is my laptop configuration:

:laptop: Acer Laptop Configuration

  • Model: Acer ALG AL15G-53
  • Processor: 13th Gen Intel® Core™ i5-13420H (8-core, 12-thread) @ 2.10 GHz
  • Graphics: Dedicated 6 GB NVIDIA GPU (likely GeForce RTX 3050)
  • RAM: 16 GB DDR4 (3200 MHz)
  • Storage: 512 GB SSD

:page_facing_up: Use Case / Workflow

My job primarily involves:

  1. Downloading scanned PDFs (public domain).
  2. Converting them into JPG images.
  3. Extracting text using OCR — mostly English, but also Indian languages like Hindi, Malayalam, Bengali, etc.

:hammer_and_wrench: What I Need

I’d like to create a virtual environment (preferably using conda or venv) that includes:

  • OCR libraries (like PaddleOCR or Tesseract)
  • Support for Indian languages
  • GPU acceleration (CUDA & cuDNN compatible with my NVIDIA GPU)
  • PDF/image tools (like pdf2image, PIL, OpenCV, etc.)

I want to avoid version conflicts and make the setup future-proof. If anyone has a similar setup or can guide me step-by-step, I’d be really grateful.

Thank you so much in advance!

Regards,
Ravi Verma

2

That is entirely impossible. The AI industry moves way too fast, and puts no value on backward compatibility.

3

Can I have an ideal workshop (virtual environment) to perform the ocr extraction work?

4

Sorry, I do not know.
“Ideal” is a very high standard.

5

Minimum possible?

6

We’ll have to wait and see if someone from the community has the information you need.

1 Like
7

Hey @adminravi , let me see if I can help you.
Your hardware configuration seems to be able to manage OCR tasks, which means work with DL or ML solutions. One point that I think that I think is good to take care of is your storage of 512 gb, which I think is too low. It is better to use an external SSD, starting with 1 TB.
Regarding the virtual environment, some time ago I created a repo of envs here that you can just download and use.
Yes, you can also use Docker, but it is a package that will eat more of your limited memory.
Hope this can help you.
Keep learning!

2 Likes
8

Thank you so much. I will be needing GPU for paddleOCR, tesseract with multiple language support enabled. I want this venv to have everything for OCR extraction from images.

9

Can you please give me the pip install path for PaddlePaddle GPU version compatible with CUDA 11.7?

10

In this link, you will find instructions on how to install CUDA 11.x to PyTorch using pip.