Real-Time Image Super-Resolution using Drone through GFPGAN and Nvidia Jetson Nano

Blind facial restoration generally uses facial priors, including a reference or a facial geometrical prior, to restore accurate and realistic details. The absence of severely low-quality inputs and the lack of high-quality references, however, significantly restrict its relevance to real-world situations. This work shows a GFPGAN model for blind face restoration that utilizes the rich and varied priors present in a pre-trained face GAN to enhance the quality and perspective of images captured by a drone in real-time. Implementing this model on an edge AI device illustrates the harmonious combination of realism and fidelity in face restoration. This is achieved by evaluating its performance on different images captured from various angles and distances. For processing on edge AI devices, the suggested system uses an Nvidia Jetson Nano, an affordable but powerful single-board computer. We present the results of a feasibility study demonstrating the capacity of GFPGAN to improve the caliber of real-time drone photography.