HTTPS SSH

CIFAR-10: Object Detection in Images

This repository contains EE660 Machine Learning course project codes.

About the dataset

CIFAR-10 is a tiny image dataset is provided by Krizhevsky and Hinton of University of Toronto [link]. This has 60,000 32x32 RBG images are divided into 50,000 train and 10,000 test images. This dataset is smaller compared to other object detection datasets, but a good one to try different machine learning algorithms.

Directory Descriptions

  • dataset: Contains the datasets.
  • keras: CNN experiments scripts.
  • notebooks: IPython notebooks of all experiments.
  • scripts: Supporting scripts
  • misc: Has some shell commands and scripts for AWS ami setup.

Key Results

Algorithm Accuracy
HOG + RF + AdaBoost 35%
HOG + SVM 42%
CNN 78%

CNN Confusion Matrix

CM

CNN Architecture

CNN_ARCH