Skip to content

Predict CIFAR-10 labels with 88% accuracy using keras.

License

Notifications You must be signed in to change notification settings

rohanchopra/cifar10

Repository files navigation

CIFAR-10 using CNNs

This project aims to predict the labels of the CIFAR-10 datset. This project uses Keras to implement deep learning. Almost all the code is in the form of IPython notebooks. Final accuracy - 87.94%

Mis-classifications

Mis-classified outputs

Metric graphs

Metrics

Dependencies

  • Jupyter
  • Keras
  • Tensorflow
  • Matplotlib
  • Pickle

Contents

  1. Helper - Helper functions which decode and fetch the data to the IPython notebooks
  2. Basic - IPython notebook to test helper functions and list images in the dataset
  3. Simple CNN - IPython notebook with a simple implementation of CNN taken from the Keras examples
  4. Improved CNN - IPython notebook which uses a pure CNN network with image augmentations to implove the accuracy of the model
  5. Model files (.h5) - Different saved models

Getting started

The quickest way to run these on a fresh Linux machine is to follow this tutorial: Kerai-Labs

Then clone this repo and start Jupyter Notebook:

git clone https://github.com/09rohanchopra/cifar10.git
cd cifar10
jupyter notebook

Tutorial

Kerai-Labs

Feedback

If you have ideas or find mistakes please leave a note.

License

MIT

About

Predict CIFAR-10 labels with 88% accuracy using keras.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published