Pytorch-UNet icon indicating copy to clipboard operation
Pytorch-UNet copied to clipboard

validation dice score is always same during training and predict generate output with 0s.

Open MNaseerSubhani opened this issue 2 years ago • 7 comments
trafficstars

I have trained the model with custom dataset, the loss gradually decreasing but the validation score always the same. I tested the image with predict.py and got 0s output with blank mask.

MNaseerSubhani avatar May 08 '23 08:05 MNaseerSubhani

i stuck in the same problem as you.

golangboy avatar Aug 07 '23 11:08 golangboy

Hi. I have the same problem. Did you manage to solve it @MNaseerSubhani ?

naguileraleal avatar Sep 24 '23 18:09 naguileraleal

我和你一样,我是自己用不同尺寸的数据集进行训练的,解决报错后能正常训练,但是DICE指标一直不变

jiao1021 avatar Nov 02 '23 12:11 jiao1021

请问你解决了吗,我也是相同的问题@jiao1021

Ashlot2002 avatar Jan 11 '24 13:01 Ashlot2002

请问你解决了吗,我也是相同的问题@jiao1021

我们修改了多次参数都无法解决。只能resize尺寸

jiao1021 avatar Jan 16 '24 13:01 jiao1021

请问你解决了吗,我也是相同的问题@jiao1021

我们修改了多次参数都无法解决。只能resize尺寸 请问你们修改后的尺寸是多少呢

AKAxiaobai-123 avatar May 20 '24 12:05 AKAxiaobai-123