pasquierjb
commented
6 years ago Nice work:
- Both CNN are rescaling the input images between 0 and 1, I changed that.
- We should use the near-infrared band of Sentinel in some way. Why not training your Sentinel model with an extra-channel?
- I noticed that we save some image with less than 5 decimals if the coordinated round up nicely. ex: 9.820000 is saved 9.82.jpg. Not a big issue but a quick fix would make it nicer.
fix the pipeline to Sentinel and Google. Custom Models. Ensambling of kNN + Ridge for remote sensing features. DB schema scripts.