Skip to yearly menu bar Skip to main content


Poster
in
Workshop: TerraBytes: Towards global datasets and models for Earth Observation

Smoothing Continual Segmentation Oscillations with Latent Domain PPCA Decoder

Marie-Ange Boum · Pierre Fournier · Dawa Derksen · Stéphane Herbin

[ ] [ Project Page ]
[ OpenReview] 
Sat 19 Jul 2 p.m. PDT — 3 p.m. PDT
 
presentation: TerraBytes: Towards global datasets and models for Earth Observation
Sat 19 Jul 9 a.m. PDT — 5:30 p.m. PDT

Abstract:

We study Domain Incremental Learning for the semantic segmentation of Earth Observation images. We demonstrate that controlling the oscillation of performance when a new domain arrives is more critical than controlling catastrophic forgetting. We propose an exemplar-free architecture that combines a large pre-trained network well adapted to dense image processing (DINOv2) and a generative decoder head based on Probabilitic Principal Component Analysis (PPCA). We validate our approach on the FLAIR#1 high-resolution dataset, which is structured as a sequence of domains.

Chat is not available.