search

HuCaoFighting / Swin-Unet

[ECCVW 2022] The codes for the work "Swin-Unet: Unet-like Pure Transformer for Medical Image Segmentation"
1.71k stars 308 forks source link

How to set hyperparameters #61

Open iaoqin opened 2 years ago

iaoqin commented 2 years ago

I would like to ask everyone, except the batch size is 24, and the popular SGD optimizer with momentum 0.9 and weight decay 1e-4. What are the other superparameter settings? The results of my own training model are much worse.

maralzar commented 2 years ago

We had the same problem please check learning size and also imagenet pretrained model which load properly. Note the learning rate written in the Git , which varies by default.