Andy Brock

Hi Qi, There can be a substantial amount of variance in the time to convergence for a model (I only had time to train one with this codebase as I...

Hmm, Looking at your logs against () this does look like it's well outside the variance I would expect. My models are also trained on 8xV100 so I don't think...

That plot looks to me like EMA isn't kicking in, or that the batchnorm stats being used alongside the EMA are stale. Can you pull the FID and IS stats...

Hi feifei-Liu, Thanks, this is helpful! Can you also post a link to your training logs, and the script you launched with? I'm working on tracking down any possible differences...

Hi Alec, You're right that cuDNN is just a speed boost-the reason I put it as a dependency is that these models can be huge (training the discriminative models from...

Hmm, I'm going to guess that's a windows 10 related bug; any chance you can try running it on a Windows 7 or a Linux machine? A quick google shows...

Most likely Theano 0.8 and Lasagne 0.2, IIRC.

IIRC (it's been a long time) those config patterns are for inference. Batch size of 50, and a double-up augmentation strategy, each epoch on a Maxwell Titan X took 6+...

The 5.8 there looks like error on the validation split (so training on 45,000 images, testing on 5,000 from the train set). If you want to use the CIFAR-10 test...

You don't need to modify the input size, just change the pooling size on [line 997](https://github.com/ajbrock/SMASH/blob/master/SMASH.py#L997) and [line 1193](https://github.com/ajbrock/SMASH/blob/master/SMASH.py#L1193) to be the appropriate size for your network. The current version...