Split Name Mismatch For TinyImageNet Loading

[h2th3k]

Here you would need to change 'validation' to 'val' to match here. Otherwise the load will fail.

[h2th3k]

@JonasGeiping Hello I wonder for TinyImageNet, what the appropriate config would be? I noticed comparing to ImageNet, InvertGrad fails to work stably. Does it need fine tunning? This is the config currently:

config = dict(signed=True,
                  boxed=True,
                  cost_fn='sim',
                  indices='top10',
                  weights='equal',
                  lr=0.1,
                  optim='adam',
                  restarts=1,
                  max_iterations=24_000,
                  total_variation=1e-6,
                  init='randn',
                  filter='median',
                  lr_decay=True,
                  scoring_choice='loss')