SEASONAL - Smart Estimation and Alteration of Scenes in Outdoor and Nature Areas using machine Learning

The goal of SEASONAL is to develop intelligent deep learning models that are able to transform outdoor images under different environmental conditions such as seasons, weather and lighting. By using state-of-the-art machine learning techniques, especially in image-to-image translation, SEASONAL will enable the seamless transformation of images between different visual domains with high quality and plausibility. One of the main challenges is to deal with different environmental changes (multi-domain) while ensuring that the generated results remain coherent and visually realistic. The project also addresses related challenges such as ensuring plausibility of transformations, improving image quality both qualitatively and subjectively by developing new no-reference quality assessment metrics to effectively evaluate the results.

 

Publications:


- Leveraging Diffusion-Based Augmentation for Robust Semantic Segmentation - ICMLA'25

METADATA

Duration: | 3 years

Start: | 2024

Funding Body: | DFG

Project Number: | 535507286