This is the official code of MICCAI23 paper "Source-Free Domain Adaptation for Medical Image Segmentation via Prototype-Anchored Feature Alignment and Contrastive Learning"
24
stars
6
forks
source link
There are some questions about training source domain model #13
When training the source domain model, I see that your config file says input_dim=3, and then the U-Net output is [batch_size，n_classes, H, W], does it mean that you input 3 slices to the network, and then only use 1 slice label to calculate the loss?
Or I would like to ask what your data looks like, is the image of each .npz file 3 slices, and then the label is 1 slice?
When training the source domain model, I see that your config file says input_dim=3, and then the U-Net output is [batch_size，n_classes, H, W], does it mean that you input 3 slices to the network, and then only use 1 slice label to calculate the loss? Or I would like to ask what your data looks like, is the image of each .npz file 3 slices, and then the label is 1 slice?