Why my SR images are black，when I trained cifar10 dataset？

[dlongry]

I trained this SRGAN project with the cifar10 dataset。I only changed the input set and generation network‘s to ：
with tf.sg_context(name='generator', act='relu', bn=True):
gen = (x_small
.sg_conv(dim=32)
.sg_conv()
.sg_conv(dim=12, act='sigmoid', bn=False)
.sg_periodic_shuffle(factor=2))
but the result is bad，the SR images are black。 and I find that the loss is always rising。I have no idea about this。Do you have meet this situation？ @buriburisuri

[DunguLock]

I meet the same problem when I trained MNIST dataset. Do you have meet this when using MNIST dataset?

[DunguLock]

@dlongry