ZurichCV #9

Linus Scheibenreif (Photogrammetry and Remote Sensing Lab, ETH Zurich): Self-Supervised Learning for Remote Sensing

Deep learning methods are commonly used for the interpretation of satellite remote sensing data. However, labeled datasets for many important remote sensing tasks are small, while large archives of unlabeled imagery exist. This motivates the use of self-supervised learning techniques in the remote sensing domain. Self-supervised learning and resolution geospatial foundation models promise to facilitate training of deep neural networks on remote sensing data and to improve performance on downstream tasks. This talk will give an overview of current approaches and challenges of self-supervised learning in remote sensing.

Pascal Chang (ETH Zurich/Disney Research): LookingGlass: Generative Anamorphoses via Laplacian Pyramid Warping

In this talk, we revisit the classical concept of anamorphosis—images that appear distorted and only reveal their true form from a specific viewpoint—through the lens of generative modeling. We present a method based on latent rectified flow models and introduce Laplacian Pyramid Warping, a frequency-aware image warping technique crucial for producing high-quality results. Our approach enables the creation of anamorphic images that not only reveal surprising hidden content when seen from the right angle but also retain a meaningful and coherent interpretation when viewed directly.