In1: a free neural network for you

At Sobolt we have decided to give back to society and make the in1 “GorGAN'' framework openly available for everyone. The name of the GAN (generative adversarial network) is derived from the Greek mythical creature Gorgon, which has snake hair and turns with her gaze anything alive to stone. One could say: ‘Okay, that's a cool name but how is that connected with what “a GAN” can do?’Each snake represents a different application - of which the framework has many. We used in1 GorGAN to artificially enhance RGBI (Red-Blue-Green-Infrared) images. For example, zooming into the picture - without the matching high resolution image actually - and removing haze.

In 2020 we began this exciting artificial data enhancement project, called in1, within the incubed program of ESA. The aim was to “see and detect more” in the 10m resolution satellite images of Sentinel-2. Part of the Copernicus mission, Sentinel-2 is a freely available satellite RGBI service of high quality images at high sampling frequency. Every 5 days the same geographic location on any place on the globe is photographed. The super resolved & haze free Sentinel-2 images support agriculture, asset tracking and urban development use cases.

Openly available
Now we took the decision to make our unique neural network - including weights and configuration files - openly available. You can deploy the in1 GorGAN for the same or other use cases and develop it further. Please note that the active development of in1 and the framework has stopped, so there is limited support for this project. Do you have the desire/need to experiment and work with a neural network for enhanced data processing and validation? Then this is for you 🙂 Check the GitHub repo.

Super resolution Sentinel-2

Comments are closed.