Night2Day - AI enhancement of night photography

Night photography often poses challenges due to reduced light availability, resulting in images with compromised details and colours. This can pose difficulty in the usability of this images across a wide range of applications (e.g. object detection, segmentation, environment description… ) or just for qualitative human judgement. In response to this, artificial intelligence (AI) algorithms designed for images can come in hand leveraging their ability of features encoding and domain translation to be able to generate a day version of a night photo. This could enlarge the usability domain of the image, not just technical but also artistic. We can just think about security cameras where a day conversion of the images would certainly help a lot the operator job without the need of using a more expensive night vision camera. In this paper, we will describe our work revolved around the implementation of an AI algorithm designed for images translation able to receive as input a night image and convert it to the counterpart daytime version while maintaining the same context and structure of the original input. The entire work has been conducted at the premises of CYENS - Centre of Excellence in Nicosia, Cyprus under the supervision of Dr. Alessandro Artusi, Xenios Milidonis and the help of the whole DeepCamera team. In this paper we will illustrate the carried out work going through all the phases from the objectives definition to the results evaluation.