Human Segmentation with Convolutional Neural Networks
Year: 2018
Type: dissertation
Abstract: In this project, a system capable of segmenting human shapes in images was developed. The first step towards that objective was a revision of the current State Of The Art (SOTA) about Deep Learning (DL), image segmentation, Convolutional Neural Networks (CNNs) and person segmentation in images. In addition, an analysis about training and testing datasets suitable for our desired task was conducted. After that, the actual implemented system will be presented. It consists of a U-Net CNN, an encoder-decoder type of network. Then, its layers and theoretical background were explained. Next, the results obtained segmenting the selected dataset and our own images were presented. Lastly, an overview of the project and the achievements as well as future lines of research finalize this Thesis.
Author: Mario Martínez Requena
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2891201766
