DSR: Towards Drone Image Super-Resolution

AbstractDespite achieving remarkable progress in recent years, single-image super-resolution methods are developed with several limitations. Specifically, they trained on fixed content domains certain degradations (whether synthetic or real). The priors learn prone to overfitting the training configuration. Therefore, generalization novel such as drone top view data, and across altitudes, is currently unknown. Nonetheless, pairing drones proper image of great value. It would enable fly higher covering larger fields view, while maintaining a high quality.To answer these questions pave way towards super-resolution, we explore this application particular focus case. We propose dataset, scenes captured at low resolutions, span altitudes. Our results show that off-the-shelf state-of-the-art networks witness significant drop performance different domain. additionally simple fine-tuning, incorporating altitude awareness into network’s architecture, both improve reconstruction (Our code data available https://github.com/IVRL/DSR).KeywordsImage super-resolutionDrone imagingSuper-resolution dataset