Learn how to classify natural scenes using deep learning techniques like normalization, augmentation, dropout, and more with ResNets and the Intel Image Classification dataset in PyTorch. Follow the step-by-step guide by John Chen in this project for Deep Learning with PyTorch: Zero to GANs.
Image Classification Using ResNets - Intel Image Classification Dataset
Author: John Chen
Welcome! This is my course project for Deep Learning with PyTorch: Zero to GANs, a course organized by freeCodeCamp and Jovian. I want to give a big shout out to Aakash, the course instructor, and the Jovian team for teaching the material and managing the course.
For this project, we will be using the Intel Image Classification dataset from Kaggle which can be found here. Original credit goes to Intel which hosted this dataset on https://datahack.analyticsvidhya.com/. The dataset contains ~ 25,000 size 150 x 150 images of six different natural scenes: buildings, forest, glacier, mountain, sea, and street. Our goal for this project is to predict the size categories with reasonable accuracy ( < 85% ).
We will also utilize a few different machine learning techniques to improve our model's performance. Some of these techniques include: normalization, augmentation, dropout, learning rate scheduling, gradient clipping, and weight decay.
Setting up the Code
Install the Jovian Python library so that I will be able to commit this notebook and its data/parameters to the Jovian website.
!pip install jovian --upgrade --quietInstall the Python library opendatasets to import and download the Kaggle dataset using its url.