search

LeeJunHyun / Image_Segmentation

Pytorch implementation of U-Net, R2U-Net, Attention U-Net, and Attention R2U-Net.
2.66k stars 594 forks source link

JS=1 and DC>1 when using ISIC2018 data #51

Closed yuheng222 closed 1 year ago

yuheng222 commented 4 years ago

Hi there,

I tried to train a model using the ISIC2018 data (2594 images with 70/10/20 split) but the JS score is always 1 and DC score is always >1.

Screenshot from 2020-02-14 22-06-18

I suspect that it might be due to my version of torch (1.2.0), torchvision (0.4.0) and pillow (6.2.0). Previously I was using torch (1.3.1), torchvision (0.4.2) and pillow (6.2.1) and my other scores were all 0 except for accuracy.

May I know what version of torch, torchvision and pillow you were using for this project? Thanks!

DeVriesMatt commented 4 years ago

Any update on this? Getting similar results

pangda72 commented 2 years ago

Hi there,

I tried to train a model using the ISIC2018 data (2594 images with 70/10/20 split) but the JS score is always 1 and DC score is always >1.

Screenshot from 2020-02-14 22-06-18

I suspect that it might be due to my version of torch (1.2.0), torchvision (0.4.0) and pillow (6.2.0). Previously I was using torch (1.3.1), torchvision (0.4.2) and pillow (6.2.1) and my other scores were all 0 except for accuracy.

May I know what version of torch, torchvision and pillow you were using for this project? Thanks! Hi,yuheng222 I got the same results too,I don't know how to solve it .Have you solved the problem yet?If it's settled, could you help me to solve it?thanks in advance.

pangda72 commented 2 years ago

Any update on this? Getting similar results

Hi,DeVriesMatt Sorry to bother you.I got similar results.Have you solved the problem yet.If it's settled, could you help me to solve it? thanks in advance.