Open pesekon2 opened 3 years ago
Currently, we have a set of deep learning models (U-Net) that we want to apply to various satellite images. We are using actinia for EO data processing, so it would be really nice to have a more generic GRASS GIS module for prediction with an existing, pre-trained model.
Do you think that detect.py
could be an OK first case for creating GRASS GIS modules for your whole project? Any thought on how to proceed here (would you welcome that kind of contributions and would you be available for review, comments and implementation of respective PRs?
For starters, I may test your docker images and try to run the project as is, so I have a better idea of how it works in general...
Do you think that
detect.py
could be an OK first case for creating GRASS GIS modules for your whole project?
Could be. The part of the code that would need to be changed is in cnn_lib.py
. The task would be more-or-less just to change the method numpy_generator to load raster maps in GRASS GIS to numpy arrays. It should be easy. But once you change it, it will work also for train.py
. We would then just need to change the UI of those two scripts to correspond with GRASS modules and skip unnecessary things like data preparation.
would you welcome that kind of contributions and would you be available for review, comments and implementation of respective PRs?
Yes and yes, yes and yes.
Actually, if you are interested in that, I could also try to spend some time on that. As there was no real interest and I didn't need it, I just postponed it and postponed, but now I could possibly try it this Fall.
For starters, I may test your docker images and try to run the project as is, so I have a better idea of how it works in general...
That makes sense. Try that and let me know. Then we'll see if it would actually be of any interest for you or not.
Would be nice one day. But the way how to do it is yet to be discussed.