Reproduce “TerraMesh-Masks: Open-Vocabulary Segmentation for Earth Observation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Benedikt Blumenstiel, Hugues Devimeux, Johannes Jakubik, Konrad Schindler

The authors introduce TerraMesh-Masks, a dataset and model adaptation for open-vocabulary segmentation of satellite imagery. They turn map geometries into language-labeled candidate masks, then train TerraMind-OVS using text queries and satellite inputs. On a 523-sample benchmark, its RGB-input positive-mask IoU is 34.96 ± 3.36%, versus 21.59 ± 2.47% for SegEarth-OV; labels mainly cover static features.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.