Why are you augmenting your support set and why is not cheat?

[brando90]

Why do you have this?

rfs/eval_fewshot.py

Line 48 in f8c837b

parser.add_argument('--n_aug_support_samples', default=5, type=int,

my impression was that in when one does an n-way, k-shot task one only uses k-shots -- but this number increases the shots. Wouldn't this be cheating?

args.n_aug_support_samples = 5

...

support_xs.size()=torch.Size([125, 3, 84, 84])
query_xs.size()=torch.Size([75, 3, 84, 84])

No aug in support set

args.n_aug_support_samples = 1

...

support_xs.size()=torch.Size([25, 3, 84, 84])
query_xs.size()=torch.Size([75, 3, 84, 84])

perhaps this is why I can't reproduce and the values reported in the paper are larger than mine -- even when I use the rfs mini imagenet checkpoint.

[brando90]

(c) during meta-testing, we create 5 augmented samples from each support image to alleviate the data insufficiency problem, and using these augmented samples to train
the linear classifier

but isn't the point to see what one can do with few shots? This puzzles me.

@WangYueFt @HobbitLong

[brando90]

@WangYueFt @HobbitLong how is increasing the support set different from increasing the shots for you?

[HobbitLong]

Hi @brando90,

If you have 5-shots, it's still 5 unique images used. n_aug_support_samples=5 means we augment each image 5 times. It's just data augmentation. The total information you used is still the same.

See table 5 in the paper. It helps a bit, but not dramatically.