🧹CleanDIFT: Diffusion Features without Noise
CVPR 2025
A novel method to extract noise-free, timestep-independent features by enabling diffusion models to work directly with clean input images.
I am currently working as a PhD student in the Computer Vision & Learning group at LMU Munich (Ommer-Lab). My research focuses on internal representations of large diffusion models. I am also passionate about applying my skills to real-world problems like bioacoustics and conservation. When I am not in front of the computer, I enjoy being outdoors, running, hiking, and scuba diving.
March 2025: Happy to start my PhD at the Ommer-Lab.
February 2025: Two papers accepted at CVPR 2025.
October 2024: First author paper accepted at WACV 2025.
July 2024: Started my internship at the Ommer-Lab.
CVPR 2025
A novel method to extract noise-free, timestep-independent features by enabling diffusion models to work directly with clean input images.
WACV 2025
We demonstrate how to distill two large vision foundation models into a smaller, high-accuracy model with lower computational cost.
Ecological Informatics
ConvNet-Transformer hybrid model enables sequence-based bat call classification, surpassing previous single call approaches..