search

modern-fortran / neural-fortran

A parallel framework for deep learning
MIT License
398 stars 82 forks source link

UNET implementation support #130

Open sidmishfpl1801 opened 1 year ago

sidmishfpl1801 commented 1 year ago

Is it possible to utilize the pre-trained Keras UNET model for making predictions? If not, can we incorporate our own custom layer instead? If neither option is feasible, could you kindly provide some suggestions for a possible workaround?

milancurcic commented 1 year ago

Currently no. I believe the key missing piece is the upsampling layer type, i.e. Keras's UpSampling2D layer. I'll open an issue for it.

milancurcic commented 1 year ago

On second thought, a transpose convolution is also missing here. So we need both transpose operations of convolutions and pooling for a UNET to work.