search

lyuzinmaxim / VUR-Net

PyTorch model VUR-Net for phase unwrapping
27 stars 5 forks source link

Give some guidance on the parameter setting #2

Open zhoulifan opened 1 year ago

zhoulifan commented 1 year ago

Hello, I have observed you mentioned the succeed train to zero cost (0.0162) at epoch 500 with Adam 0.0001 very fast. Howerver, I cannot obtain the loss as low as you said. I used the automatic generated synthetic data provided by dataset_generation.py. My question is how many phase images are gernerated? My accuracy of unwrapping cannot attain the 50% above. I am hoping you could give me some guidance on the parameter setting.

lyuzinmaxim commented 1 year ago

Hi!

I can't say exactly, but it was like several thousands of images, which were synthetically generated (here was 50 just for testing) https://github.com/lyuzinmaxim/VUR-Net/blob/59168a529fe879daeac8fa2f03ea60b91ea4e0b3/main.py#L9