tensorflow-deeplab-resnet icon indicating copy to clipboard operation
tensorflow-deeplab-resnet copied to clipboard

Published 20 hours ago • verified publisher icon DrSleep

Mean IoU of 0.008, even when training to overfit.

Open ghost opened this issue 6 years ago • 1 comments

I use my own dataset, but I train using just one image. The loss goes down to as low as 0.004. But when I evaluate using the same image I use for training, I get a mean IoU of 0.008. Why could this be the case? I made sure my image and label was being read correctly.

ghost avatar Aug 15 '18 01:08 ghost

I have a same issue the loss is going to a small value,but the mean IOU is very bad.

Simpatech-app avatar Apr 04 '19 21:04 Simpatech-app