While the second epoch of my CycleGAN model was running I ran into this error
RuntimeError: cuDNN error: CUDNN_STATUS_BAD_PARAM
and the error pointed to this function
def pass_through_discriminator(discriminator, image):
score, k = 0, Variable(torch.zeros(1)).type(dtype)
xp, yp = 0, 0
x, y = 70, 70
offset = 25
while x < 128:
while y < 128:
k += 1
score += discriminator(image[:, :, xp:x, yp:y])
yp += offset
y += offset
xp += offset
x += offset
return score / k
I read in other topics that the size of the image has to be increased, but am not very clear.
Any suggestions on how to solve this
Thanks in advance !