efficientnet-b8 and AdvProp

[seefun]

With advprop, efficientnet got greater score in ImageNet. Would you update to the new ckpt?
the paper: https://arxiv.org/pdf/1911.09665.pdf
https://github.com/tensorflow/tpu/tree/master/models/official/efficientnet

[lukemelas]

Awesome, thanks for posting this. It's on the way.

[Oktai15]

@lukemelas maybe your current code in “tf_to_pytorch” dir will work?

[laurentdillard]

@lukemelas any update on that ?

[jionie]

@lukemelas any update on that ?

[seefun]

[lukemelas]

Apologies for the delay on this (I had final exams this past week). Coming soon.

[laurentdillard]

Hi @lukemelas , just wondering if you had the chance to work on that one.

[lukemelas]

Sorry this took forever. It should be in now :)

Let me know if you have any issues.

[lukemelas]

Closing this, but feel free to re-open it if you have any issues/questions.

[sliawatimena]

Dear all,

I am using https://colab.research.google.com/drive/1Jw28xZ1NJq4Cja4jLe6tJ6_F5lCzElb4
Why the efficientnet-b0 advprop=True result is very bad?
Thank you.

Suryadi

Loaded pretrained weights for efficientnet-b0, advprop=False
-----
giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca           (83.44%)
brown bear, bruin, Ursus arctos                                             (0.62%)
lesser panda, red panda, panda, bear cat, cat bear, Ailurus fulgens         (0.60%)
ice bear, polar bear, Ursus Maritimus, Thalarctos maritimus                 (0.44%)
Arctic fox, white fox, Alopex lagopus                                       (0.34%)

Loaded pretrained weights for efficientnet-b0, advprop=True
-----
wombat                                                                      (3.34%)
candle, taper, wax light                                                    (2.00%)
Angora, Angora rabbit                                                       (1.87%)
schipperke                                                                  (1.86%)
hog, pig, grunter, squealer, Sus scrofa                                     (1.60%)

[lukemelas]

Did you use the advprop image preprocessing or the usual preprocessing? See https://github.com/lukemelas/EfficientNet-PyTorch/blob/master/examples/imagenet/main.py#L211. That's the reason advprop is not enabled by default. Let me know if it still doesn't work and I can look into it.