Andy Brock

Just drop a FashionMNIST data-loader in [utils.py](https://github.com/ajbrock/FreezeOut/blob/master/utils.py#L71) and replace the mean / average statistics. Run it with the options that turn FreezeOut off, something like: `python train.py --t_0 1.0 --epochs...

Probably try one of those shake-shake or shake-drop variants, with a well-tuned SGDR cycle and distillation a-la born again neural nets or stochastic weighted averaging, and figure out the right...

Hah, I'm not saying that datasets that haven't been hit with the latest-and-greatest aren't properly benchmarked, but no one's beaten a vanilla WRN40-4 with standard data aug. Kagglers would have...

If you remove all my explicit conv2DDNNLayer calls (aliased as C2D) and replace them with standard Conv2D layer calls it should work, but keep in mind that those DeconvLayers are...

I've added support for not explicitly making DNN calls, so you should, presumably, be able to try running this on CPU. At the very least, it shouldn't crash.

@dkandalov Change the dnn=True line on line 53 to dnn=False, and it won't attempt to import cuDNN.

Yes, following a commit after the above comments, the proper line for this is now 18.

There's plenty that can go wrong. Just based on these images it looks like even the VAE half of the network isn't working. I'd recommend starting by training DCGAN in...

Are you making sure to not propagate reconstruction gradients to the discriminator? I've always kept the "encoder" as a small MLP (or even a single dense layer) that operates on...

I've not used Docker at all (well, I've not *successfully* used Docker at all), but I can try to point you in the right direction. I was running this with...