Deep Learning model for Polarization

While polarimetric imaging can provide shape and surface information about objects in a scene, it does not provide full three-dimensional (3D) shape information. However, when additional information is provided about the imaged scene, it has been shown that 3D scene reconstruction is possible, known as structure from polarization (SfP). The goal of this research is to investigate deep learning approaches to the SfP problem. As part of this research, visible polarimetric imagers will be used to build a set of training data of various geometric objects that can be easily modeled in a 3D software environment. Full 3D image information will then be generated and registered with collected polarimetric data as part of the image training set. Deep learning architectures will then be investigated and trained and performance evaluated to determine reconstruction quality.

We can share the dataset to work on it. We will need MATLAB code and detailed report.