robgrzel Goto Github PK

followers: 2.0 following: 5.0 repos: 116.0 gists: 18.0

Name: Robert Grzelka

Type: User

Location: Sweden, Gotheborg

Robert Grzelka's Projects

an-example-of-cnn-on-mnist-dataset

An Example of CNN on MNIST dataset

This is my Thesis. In this project we implemented and studied an Animal Detection System, aiming at both high detection accuracy and computational efficiency. Specifically, we propose a two-stage detection method. The reason we employ two machine learning algorithms is the reduction of the false-positive rate and the execution time required for the multi-scale detection procedure of an input image. In the first stage, we use the Histogram of Oriented Gradient (HOG) [1] descriptor and a Support Vector Machine (SVM) [2] classifier that provides the next stage with a set of regions of interest (ROI) containing target animals and other false positive targets. Subsequently, the second stage, rejects the false positive ROIs by using a Convolutional Neural Network (CNN) [3] classifier. To train and evaluate the animal detector, we use CIFAR-10 dataset [7]. The purpose of this method was to detect animals from images and videos captured in natural scenes. For our experiments we used deers as the animal of choice.

applying_eanns

A 2D Unity simulation in which cars learn to navigate themselves through different courses. The cars are steered by a feedforward neural network. The weights of the network are trained using a modified genetic algorithm.