jpowie01 / cifar10-handson Goto Github PK

This hands-on was prepared for one of my presentations that took place on Computer Vision's mini-course at student's orgranization called "Gradient" (Gdańsk University of Technology).

Before running this hands-on, please recall the materials from previous meetings, especially my "Introduction to CNNs", which you can find in the root directory of the repository. HTML version with all the GIFs is available here.

What do you need to install to run this hands-on?

• Linux/macOS,
• Python 3.5+,
• Virtualenv (nice to have).

Note: PyTorch doesn't officially support Windows! For more information how to use it on Windows, please refer to this Issue on GitHub. It's highly recommended to use Linux/macOS instead, even in the Virtual Machine.

Steps to follow:

1. Install all needed packages...

• ...into local environment (in order not to mess with global Python packages):

$ virtualenv -p python3 venv
$ . venv/bin/activate
(venv) $ pip install -r requirements.txt

• ...OR into global Python packages:

$ pip3 install -r requirements.txt

2. Install PyTorch (it's needed to do separately):

• Open official PyTorch documentation,
• In the section "Get Started" select your Operating System,
• Choose "pip" as your Package Manager,
• Select Python version which you use (python --version),
• Select CUDA version (not needed for this hands-on),
• Run commands that will show you below the form.

Now, you're ready to go!

Once your environment is ready, it's time to open Jupyter Notebook with the hands-on:

(venv) $ venv/bin/jupyter notebook

...or if you installed the packages globally:

$ jupyter notebook

Once your browser will open by itself, select the notebook called CIFAR10 Hands-on.ipynb and follow the instructions inside of it. There is also a separate notebook with all completed steps called CIFAR10 Hands-on [COMPLETED].ipynb. Don't hesitate to look into this one if you want to check the answers :)

IMPORTANT: Please run first two code blocks before the meeting! These will check your environments for installed packages and download the CIFAR10 dataset (~160MB) from the Internet. Internet connection on our university may be weak, so do this prior the meeting!