Cnn model doubt

Hi,

My model using UNet is showing a very uneven result after running all the epochs.

But after running, again and again, all the epochs it is now very stable.

So, the next day when again I run the model it again shows extremely bad uneven results.

What is the problem behind it?

please give me a solution

Using Unet,
MSE loss is okay where testset loss is greater than trainingset loss,
but for binary cross-entropy loss trainingset loss is 0.1 greater than test set loss, is it overfiting?

What is the problem statements? what is the datasize?
Can you Kindly show the graphs or numbers MSE loss for understanding better?
Are you using keras for this?