Detekce reálných objektů podle syntetických předloh

Abstract

The purpose of this diploma thesis is to develop and test software application for finding objects in images based on synthetic pattern. Resulting application will be able to guess position of object in the image and guess harsh estimate of its rotation for user. I made research a tested several neural networks for solving this problem. I stared with Unitbox and continued with Darknet implementation of YOLO and finally with TensorFlow YOLO implementation and own implementation. I managed to create network that satisfies the assignment.