Closed rayanelleuch closed 7 years ago
Hi Rayan,
You are correct, it seems ordinary dropout was used instead of spatial dropout due to a mistake in the code :confused:.
Ok thanks! I try to reproduce it on tensorflow but still kind of overfitting a lot from my point of view with spatial dropout. I will try with the normal dropout.
On the paper released it says that SpatialDropout was used in the Unet model. But in the current code I just see the use of normal DropoutLayer. I saw that there is the implementation of SpatialDropout in custom_layers.py but I don't see it used. Is the code exactly the same used for Luna challenge?