U-Net approximating the version described in Khallaghi et al (2025), but with a stock VGG19 backbone and pre-trained ImageNet weights. The model is trained to recognize 3-classes (non-field, field interior, field edge) using the Planet imagery-based Mapping Africa catalog. See here for code and configuration.

Performance metrics on validation split:

pixel_level_iou 0.5788
pixel_level_precision: 0.7310
pixel_level_recall: 0.7354
object_level_precision: 0.2781
object_level_recall: 0.2665

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support