DSLR Quality Photos for Mobile Devices Using Deep Learning

The proliferation of the smartphone in the modern era brings with it a strong need to improve the quality of the camera accompanying the mobile phone. The very nature of the mobile phone forces physical limitations on the camera's optical system, such as small sensors, compact lenses and limited hardware. These restrictions prevent phone cameras from reaching a quality comparable to professional cameras.
In our project, we worked with an end-to-end network that aims to study the transformation from poor quality image space to high quality image space. This is achieved by minimizing a number of loss functions, each of which takes care of a particular aspect of image quality.
Our main goal is to improve network results and we tried to do this in a number of ways, such as switching color space from RGB to CIE-Lab, data pre-processing, upgrading the loss function, mask learning, changing network architecture and using L1 norm.
For each enhancement attempt we drew conclusions from the results obtained and studied its limitations to the image enhancement problem. In some of the attempts we were unable to show improvement and we even got a decline in performance, while in others we showed improvement in both the objective metrics and our own subjective evaluation.

DSLR Quality Photos for Mobile Devices Using Deep Learning
DSLR Quality Photos for Mobile Devices Using Deep Learning