Model weights from the release at https://github.com/fieldsoftheworld/ftw-baselines/releases/tag/Pretrained-Models
The "full" models are trained on all country data in the Fields of The World dataset, while the CCBY models are trained according to the subset described here.
Example usage:
import segmentation_models_pytorch as smp
import torch
model = smp.Unet(
encoder_name="efficientnet-b3",
encoder_weights=None,
in_channels=8,
classes=2
)
model.load_state_dict(torch.load("ftw-2class-full_unet-efficientnetb3_rgbnir_f2444768.pth", weights_only=True))