In [ ]:
%reload_ext autoreload
%autoreload 2
from fastai.vision import *
Dogs and cats¶
Resnet 34¶
In [ ]:
path = untar_data(URLs.DOGS)
path
Out[ ]:
PosixPath('/home/ubuntu/.fastai/data/dogscats')
In [ ]:
data = ImageDataBunch.from_folder(path, ds_tfms=get_transforms(), size=224).normalize(imagenet_stats)
data.show_batch(rows=4)
In [ ]:
learn = cnn_learner(data, models.resnet34, metrics=accuracy)
learn.fit_one_cycle(1)
Total time: 02:54
| epoch | train_loss | valid_loss | accuracy |
|---|---|---|---|
| 1 | 0.050182 | 0.028986 | 0.990000 |
In [ ]:
learn.unfreeze()
learn.fit_one_cycle(6, slice(1e-5,3e-4), pct_start=0.05)
Total time: 07:19
| epoch | train_loss | valid_loss | accuracy |
|---|---|---|---|
| 1 | 0.048187 | 0.027477 | 0.988500 |
| 2 | 0.022563 | 0.024907 | 0.990500 |
| 3 | 0.017952 | 0.017247 | 0.995000 |
| 4 | 0.016476 | 0.018968 | 0.993500 |
| 5 | 0.008154 | 0.019849 | 0.992000 |
| 6 | 0.009654 | 0.018833 | 0.993500 |
In [ ]:
accuracy(*learn.TTA())
Total time: 02:09
Out[ ]:
tensor(0.9945)
rn50¶
In [ ]:
learn = cnn_learner(data, models.resnet50, metrics=accuracy)
learn.fit_one_cycle(1)
Total time: 01:17
| epoch | train_loss | valid_loss | accuracy |
|---|---|---|---|
| 1 | 0.044986 | 0.028825 | 0.989000 |
In [ ]:
learn.unfreeze()
learn.fit_one_cycle(6, slice(1e-5,3e-4), pct_start=0.05)
Total time: 08:13
| epoch | train_loss | valid_loss | accuracy |
|---|---|---|---|
| 1 | 0.040551 | 0.021553 | 0.991500 |
| 2 | 0.028230 | 0.019264 | 0.992000 |
| 3 | 0.024131 | 0.023275 | 0.991500 |
| 4 | 0.010579 | 0.020088 | 0.993000 |
| 5 | 0.007140 | 0.018935 | 0.993000 |
| 6 | 0.005668 | 0.019965 | 0.993000 |
In [ ]:
accuracy(*learn.TTA())
Total time: 01:04
Out[ ]:
tensor(0.9940)