[Solved] I think this happens because I pass gt image as original image and original image as gt image.
After I fix it, suddenly getting empty tensor error has disappeared
---------------------------------------------------------------------------
Output was being generated normally
...
[[0.3303, 0.3251, 0.3482, ..., 0.3575, 0.3505, 0.3299],
[0.3292, 0.3334, 0.3629, ..., 0.3746, 0.3608, 0.3424],
[0.3399, 0.3466, 0.3729, ..., 0.3780, 0.3643, 0.3420],
...,
[0.2718, 0.2642, 0.3199, ..., 0.3800, 0.3671, 0.3445],
[0.2831, 0.2769, 0.3199, ..., 0.3685, 0.3552, 0.3367],
[0.2862, 0.2885, 0.3112, ..., 0.3392, 0.3283, 0.3198]]],
...
Suddenly, I got emply tensor
tensor([], device='cuda:0')
I used rather large learning rate like 1.0, 0.1
Does this error come from too large learning rate or is there other potential reasons?