10,000 Crop Field Boundaries across India

Release of dataset and neural network weights accompanying the paper "Unlocking large-scale crop field delineation in smallholder farming systems with transfer learning and weak supervision" (forthcoming in <em>Remote Sensing)</em>. Ten thousand crop fields in India were delineated manually through inspection of high-resolution satellite imagery (Airbus SPOT). We also provide the weights of the highest performing neural network (FracTAL ResUNet architecture) pre-trained in France and fine-tuned on Airbus SPOT images in India. The model was trained in MXNet 1.6.0 and can be loaded with the "model.load_parameters()" function.