Subimage Overlap Prediction: Task-Aligned Self-Supervised Pretraining For Semantic Segmentation In Remote Sensing Imagery

Lakshay Sharma, Alex Marin · 2026

Read Subimage Overlap Prediction: Task-Aligned Self-Supervised Pretraining For Semantic Segmentation In Remote Sensing Imagery by Lakshay Sharma, Alex Marin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-01781.txt.

Description

Satellite image analysis needs labeled examples, and self-supervised pretraining can reduce that need, but most methods demand huge image collections. Lakshay Sharma and Alex Marin propose a leaner pretraining task for remote sensing: cut a small patch from an image and train the model to paint exactly where that patch came from.

Called subimage overlap prediction, the task forces the model to learn spatial layout before it ever sees a label. Models pretrained this way converged faster and matched or beat alternatives on downstream segmentation across several architectures and datasets, with the advantage growing as labeled data shrank. The method also needed far less pretraining imagery than competing self-supervised approaches, and the code and weights are released.

Catalog record

CREATORLakshay Sharma, Alex Marin
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-01781

Browse the full library · About the Tower of Babel