Sentinel-hub seems to be a really great platform as said “an engine for processing of petabytes of satellite data. It is opening the doors for machine learning”.
But I am confused about the availability of a feature. as we all know, first step for any machine learning algorithm is training using ground truth data. depending on the complexity of the model this step can be too much resource-consuming.
now which one is true?
1 - it is possible to handle this step using sentinel-hub
2 - it is not possible to handle training in sentinel-hub and it is necessary to download the needed satellite data and do training locally or anywhere else. after training the model you can convert it to an evalscript and use for any other region and time.
3 - 2 is correct. but you can not convert(stringify) any machine learning model. some simple models like decision tree are possible. but deep learning models is not possible or too much prune to errors for conversion.
I thank in advance can explain it.