This repo contains a pytorch implementation of a VGG19 trained on the Sentinel-2 classification dataset EuroSat
Performing classification on this dataset is in now way shape or form a hard task. This was just done to produce a set of VGG19 weights, susceptible for a four channel input and kernels that are trained to detect features in satellite imagery.
Those weights can be used to serve as a perceptual loss for other remote sensing machine learning tasks, or a p retrained network for fine tuning if you don't have a GPU.
Here a peak view on how training process usually (with default parameter) looks over 50 epochs:
