Garnot Vivien Sainte Fare ; Landrieu Loic

Abstract

Unprecedented access to multi-temporal satellite imagery has opened newperspectives for a variety of Earth observation tasks. Among them,pixel-precise panoptic segmentation of agricultural parcels has major economicand environmental implications. While researchers have explored this problemfor single images, we argue that the complex temporal patterns of cropphenology are better addressed with temporal sequences of images. In thispaper, we present the first end-to-end, single-stage method for panopticsegmentation of Satellite Image Time Series (SITS). This module can be combinedwith our novel image sequence encoding network which relies on temporalself-attention to extract rich and adaptive multi-scale spatio-temporalfeatures. We also introduce PASTIS, the first open-access SITS dataset withpanoptic annotations. We demonstrate the superiority of our encoder forsemantic segmentation against multiple competing architectures, and set up thefirst state-of-the-art of panoptic segmentation of SITS. Our implementation andPASTIS are publicly available.