/neural_networks

Primary LanguageJupyter Notebook

Neural Networks

This repo is a collection of simple image classifiers using neural networks.

Table of Contents:

Technologies

  • Python v3.x
  • Tensorflow v2.x or Tensorflow-GPU v2.x
  • PyTorch v1.5.x
  • OpenCV v4.3.x
  • Numpy
  • Matplotlib

Installation of libraries:

Tensorflow:

pip install tensorflow

NOTE: If your system has a CUDA compatible GPU (Nvidia cards):

pip install tensorflow-gpu

CIFAR10

The CIFAR dataset is an enormous collection of images belonging to ten diferent classes.


cifar-img

Results:

Network Architecture

I have used Tensorflow and Keras here to create the neural networks. There are three sections to the project:

  1. A simple Neural Network that produces the least accurate results. This serves as teh benchmark for the rest of the project.
  2. A simple Convolutional Neural Network.
  3. Hyperparameter tuned CNN which produces the best accuracy.

Dataset

Dogs or Cats?

The program aims to distinguish between dogs and cats in an image.