Hloubkové mapy pomocí Deep Learningu

Abstract

This thesis focuses on the depth map generation using deep learning for stereo images. For implementation, convolutional neural network in a form of an autoencoder is used together with a bilinear transformation. Siamese method is being observed as the main topic in this work. Implementation is written in Python and PyTorch, which uses CUDA at it’s backend optionally for speeding up the learning process.