README.md

Computational Active Depth from Defocus

My Master’s thesis focuses on 3D reconstruction technologies. I am building a novel active depth sensing system that infers depth by analyzing the blurriness of the projection pattern at different depth levels caused by camera defocus. Without relying on complex hardware, I constructed multiple parametric (Gaussian based) and non-parametric (DNN-based) computational models for depth inference. This repository contains a number of different models implemented for my thesis work, more info can be found on my page.

Parametric computational model (Gaussian-based):

Circularly-Symmetric 2D Gaussian [paper]

Elliptical 2D Gaussian: [paper]

Non-parametric computational model (DNN-based):

The DfD_matlabNN folder contains a collection of models that are inplemented using the native Matlab Neural Network Toolbox. [paper]

The DfD_tf folder contains a collection of models that use TensorFlow's high-level APIs. [paper]