Hi Deep Learners,
We are happy to announce the next Vienna Deep Learning Meetup on November 25 at Fiskaly. Our Agenda:
\\\
Agenda:
18:15 Arrival
18:30 Welcome by the meetup organizers
Introduction by the hosts
18:45 Talk 1: From Metadata to Labelled Imagery: MaskGIT for Land Cover Synthesis and ControlNet-Guided Diffusion for Earth Observation Images by Lydia Abady and Sebastian Ratzenböck (Another Earth)
19:30 Announcements
Networking Break
20:00 Talk 2: Inside TED AI by Jakob Reiter (Mimirio)
20:30 Networking
\~22:00 Wrap up & End
\\\
Talk Details:
\-\-\-\-\-\-\-\-\-\-\-\-\-\-
Talk 1: From Metadata to Labelled Imagery: MaskGIT for Land Cover Synthesis and ControlNet-Guided Diffusion for Earth Observation Images
Pixel-level labels for Earth observation are expensive to produce and unevenly available across regions, classes, and resolutions. What if we could generate the labels first and then the images to match them? In this talk, we present a two-stage approach to generating synthetic image-label pairs. First, a MaskGIT-style masked token transformer synthesizes land cover maps conditioned on metadata such as class distribution, ecoregion, city, and ground sample distance. These conditions guide the composition and geographic characteristics of the generated scenes.
Second, a diffusion model generates corresponding optical imagery, using ControlNet to incorporate the land cover map alongside metadata conditioning. Because the image is generated from the map, every sample comes with labels aligned pixel by pixel.
Through a visual walkthrough of generated land cover maps and their matching images, we show how changing the metadata reshapes a scene, how the same layout looks across different regions and resolutions, and how this approach can produce labelled data on demand for rare classes and under-represented areas.
About the speakers:
Lydia Abady is an AI Engineer and researcher specializing in generative AI, machine learning, and computer vision. S