prithvi-landslide

Fine-tuned Prithvi-EO-V2-300M (Prithvi-EO-2.0 backbone, IBM/NASA) checkpoint for landslide.

Binary landslide segmentation, fine-tuned on Landslide4Sense.

• Number of output classes: 2
• Trained with TerraTorch / PyTorch Lightning
• last.ckpt -- Lightning checkpoint (includes optimizer state; strip with torch.load(..., map_location="cpu")["state_dict"] if you only need weights)
• config.yaml -- full resolved training config for this run (model args, data module, trainer settings)

This is a personal fine-tune, not an official IBM/NASA release. Base model license: MIT (see ibm-nasa-geospatial/Prithvi-EO-2.0) -- verify current upstream license/terms before downstream use.