In [ ]:

%reload_ext autoreload
%autoreload 2

from fastai.vision import *
from fastai.vision.models.wrn import wrn_22

torch.backends.cudnn.benchmark = True

In [ ]:

path = untar_data(URLs.CIFAR)
path

Out[ ]:

PosixPath('/home/ubuntu/.fastai/data/cifar10')

In [ ]:

ds_tfms = ([*rand_pad(4, 32), flip_lr(p=0.5)], [])
data = ImageDataBunch.from_folder(path, valid='test', ds_tfms=ds_tfms, bs=512).normalize(cifar_stats)

In [ ]:

learn = Learner(data, wrn_22(), metrics=accuracy).to_fp16()
learn.fit_one_cycle(30, 3e-3, wd=0.4, div_factor=10, pct_start=0.5)

Total time: 07:08

epoch	train_loss	valid_loss	accuracy
1	1.493729	1.288911	0.532400
2	1.161237	1.103286	0.604100
3	0.958353	0.996172	0.649300
4	0.829848	1.120279	0.638500
5	0.716744	0.724809	0.752300
6	0.634061	1.139240	0.626800
7	0.574845	1.627489	0.506100
8	0.531848	0.912567	0.712200
9	0.489270	0.791987	0.745500
10	0.459794	0.646239	0.782000
11	0.431601	0.640238	0.789400
12	0.402780	0.648663	0.793200
13	0.387314	0.614063	0.793000
14	0.366800	0.594612	0.813600
15	0.338351	0.620742	0.804600
16	0.324927	0.470762	0.841500
17	0.302258	0.468217	0.844900
18	0.286116	0.421791	0.859000
19	0.257466	0.428825	0.859200
20	0.233121	0.343100	0.887100
21	0.205734	0.342273	0.887500
22	0.176312	0.318532	0.896700
23	0.144774	0.328396	0.896100
24	0.119999	0.287829	0.910800
25	0.087010	0.232755	0.928600
26	0.060723	0.236310	0.931400
27	0.042571	0.207955	0.942000
28	0.027802	0.217585	0.938900
29	0.020010	0.209865	0.943000
30	0.016231	0.209546	0.943000

With mixup

In [ ]:

learn = Learner(data, wrn_22(), metrics=accuracy).to_fp16().mixup()
learn.fit_one_cycle(24, 3e-3, wd=0.2, div_factor=10, pct_start=0.5)

Total time: 05:42

epoch	train_loss	valid_loss	accuracy
1	1.806122	1.413667	0.504500
2	1.592665	1.189260	0.590700
3	1.461559	1.018693	0.655400
4	1.370229	0.874307	0.712100
5	1.296808	0.913873	0.704000
6	1.250895	0.836409	0.733900
7	1.209640	0.736776	0.778600
8	1.186605	0.753798	0.767200
9	1.166516	0.757842	0.767700
10	1.137516	0.699450	0.806500
11	1.120571	0.736078	0.780600
12	1.103785	0.909942	0.710700
13	1.073971	0.530825	0.856600
14	1.055455	0.583879	0.831600
15	1.035860	0.509721	0.868300
16	1.017207	0.510995	0.867800
17	0.995223	0.446647	0.889100
18	0.962532	0.378901	0.904300
19	0.940812	0.352570	0.917800
20	0.922071	0.332144	0.928500
21	0.899262	0.326830	0.932000
22	0.880337	0.312892	0.936600
23	0.874789	0.306469	0.940000
24	0.865873	0.305611	0.939200