How do you extend unet_learner to more than 3 input channels?

(Nicholas Pilkington) #1

I have 6 channel input imagery that I’m trying to do semantic segmentation on. I’ve extended the SegmentationItemList to load the imagery. But how do I extend the unet_learner to handle 6 channel images as input?

I’ve looked on the forum and there are 2 topics but neither give a conclusive answer.


(Navid Panchi) #2

Check out this Github Repo out. I have defined a modified unet_learner. It will help you out.

1 Like

Train a model 4 or more input channels with pretrained weights