hanban

Thx Mokke. I totally understand your points. I found something kinda interesting in Glow source code, which is they were not using affine coupling layer instead they used additive coupling....

Thx again for explaining all these! There is one more thing I found in your code recently. For the sake of convenience, I post it here, hope you don't mind....

@MokkeMeguru Hey, this issue make me curious how does tfp version of glow generating celeba look like... so I tinker your code a bit, adding variational dequantization and making neural...

 downloaded from google drive and found out the alignment is different from FFHQ-style and images discribed in this repo readme.

> Codes will be available at https://github.com/TencentARC/FastRealVSR. What a joke, Tencent! XD

@levimcclenny Thanks for your suggestion! I did `train_dataset = strategy.experimental_distribute_dataset(train_dataset)` but I forgot to put it in my code. I also try remove randomness from map function by removing `tf.image.random_brightness(img,...

Some info from profiler, it might be helpful?  

@wujiren So you think `@tf.function` slows things down? I don't think removing `tf.function` is a good idea tho. @nikitamaia Here's my trace view for one loop of training. Idk if...

@nikitamaia any suggestion?

`tf.distribute.experimental_set_strategy(strategy)` , `strategy.scope()` and `tf.distribute.ReductionToOneDevice` are used. Doesn't seems Single GPU:  Two GPU:  ![Screen...