Deep Learning Quick Reference

Model evaluation

The CIFAR-10 dataset is certainly more complicated and the network has quite a few more parameters. As such, things are going to take longer. Here's what our images look like on epoch 0, batch 300:

I'm starting to maybe see some edges, but it doesn't really look like anything. If we wait a few epochs though, we're clearly in fuzzy squirrel and weird fish territory. We can see something taking shape, it's just all a little fuzzy, as shown in the following image:

The following image shows our generator after 12 epochs:

>

I see ...