Projektdaten

---

SEASONAL - Intelligente Schätzung und Veränderung von Szenen im Freien und in der Natur durch maschinelles Lernen

---

The goal of SEASONAL is the transformation of outdoor images between different environmental conditions (e.g., season, weather, and lighting) with high quality and plausibility. As an example of such changing conditions, Figure 1.1 shows some of the changes a natural scene can undergo during the course of multiple seasons. The transformation of an image from one visual domain to another is usually referred to as lmage-to-image translation. lt is a sub-problem of the general image generation task and is commonly addressed through generative models that analyze !arge amounts of data and replicate the underlying causal relations. The introduction of neural networks and deep learning has sparked significant developments in this area within the last 10 years. Still, none of the currently existing approaches can translate images across the various environmental conditions we aim to address in this project. The challenges that are connected to this task primarily concern the ability to handle multiple domains while generating high-quality output. In addition, we want to focus on the plausibility of the results, an aspect often overlooked in current image generation approaches. In the next section, we give a brief overview of currently existing generative models. Afterward, we describe the evolution of the current state-of-the-art in image-to-image translation. In the last section, we present an overview of the existing measures for the automatic assessment of images.