I'm trying to use Contrastive Learning or Self Supervised Learning to segment 2D medical images.
I want to use something like SimCLR or SimSiam, however, I'm getting stuck on how it should work (for example using this code).
What should be my approach when training SimSiam with a U-Net like backbone?
Thank you!
this is the possible approach
approach 1: training a model with U-NET and doing the segmentations and predicting the medical diseases with the help of segmentation. This is a supervised approach.
approach 2:This is semi-supervised approach. Training a dataset with U-NET for contrastive learning. so you get meaningful features in your latent space and then you train your normal segmentation u-Net while using the pre-trained Encoder part of your contrastive learning U-Net. Here to remember this normal segmentation u-net would be different from the U-NET for contrastive learning. So, there would be 2 different classes for U-NET. and then simply do the predictions by the semantic segmentation approach.
