|Betreuer:||M.Sc. Martin Pöllot (Raum 01.178)|
|Hochschullehrer:||Prof. Dr.-Ing. André Kaup|
Convolutional Neural Networks (CNNs) represent the best tool for classification of image content. Over the past few years, this area of research brought great progress to image classification. One of the most significant breakthroughs in the beginning is the reliant classification of handwritten postal zip numbers and later the recognition of faces or license plates. Current state of the art applications use powerful real-time-capable networks that are able to detect multiple classes in images for detecting pedestrians, vehicles, obstacles and traffic signs in real-time. A CNN is rated by its overall ability to classify its input. At the moment there is a plethora of data sets with varying number of classes, training, validation and test images. Many of these data sets are arranged in different ways, which makes handling them difficult and tedious.
The thesis shall consider the acquisition and arrangement of a wide variety of datasets with standardized loading functions. These functions should be implemented in such a way, that regardless of the specific dataset, dimensions, color representation like RGB or BGR will all be the same. Furthermore, a list with an overview of all acquired datasets should be composed showing information about the most significant aspects of each dataset. The main part consists of the implementation of state of the art convolutional neural networks and compare the networks on different datasets with different hyper-parameters.