This repository contains three fine-tuned TerraMind models, using terramind_v1_base backbone. They are able to segment respectively seismic fault scarps, paleo-landslides, and river active channels, based on Sentinel-2 RGB and infrared images as well as a Digital Elevation Model (DEM). A UPerNet decoder is added to the TerraMind backbone, and the fine-tuning is done on three released datasets, one for each object (c.f. GitLab repository).
- Full repository: GitLab.
- Seismic fault scarp fine-tuned model available here: TerraMindUPerNet_sfs.pth
- Paleo-landslide fine-tuned model available here: TerraMindUPerNet_pls.pth
- Active river channel fine-tuned model available here: TerraMindUPerNet_arc.pth