This repository contains a project on using monocular distance estimation with neural networks to create a portrait photography effect.
The task is divided into two parts: segmentation of a person in a photo and creation of a depth map. To create more smooth transitions between different image layers, Gaussian filters are used. Below is an examples of how the different parts of the pipeline work.
Technical details on work are availabe in media/Report.pdf file
Example of models' work result
Using the models without any smoothing cause many artifats. Thus we apply Gaussian filtering:
