Review: computer vision transfer learning¶

This is a subset of lesson 1 of https://course.fast.ai

In [1]:

%reload_ext autoreload
%autoreload 2
%matplotlib inline

from fastai.vision import *
from fastai.metrics import error_rate

In [2]:

bs = 64
# bs = 16   # uncomment this line if you run out of memory even after clicking Kernel->Restart

We are going to use the Oxford-IIIT Pet Dataset by O. M. Parkhi et al., 2012 which features 12 cat breeds and 25 dogs breeds. Our model will need to learn to differentiate between these 37 distinct categories. According to their paper, the best accuracy they could get in 2012 was 59.21%.

In [5]:

path = untar_data(URLs.PETS)
path_anno = path/'annotations'
path_img = path/'images'
np.random.seed(2)
fnames = get_image_files(path_img)
fnames[0]

Out[5]:

PosixPath('/home/jhoward/.fastai/data/oxford-iiit-pet/images/great_pyrenees_173.jpg')

In [6]:

pat = r'/([^/]+)_\d+.jpg$'
data = ImageDataBunch.from_name_re(path_img, fnames, pat, ds_tfms=get_transforms(), size=224, bs=bs
                                  ).normalize(imagenet_stats)
data.show_batch(rows=3, figsize=(7,6))

Training¶

In [9]:

learn = cnn_learner(data, models.resnet34, metrics=error_rate)

In [10]:

learn.fit_one_cycle(4)

epoch	train_loss	valid_loss	error_rate	time
0	1.411342	0.298996	0.086604	00:15
1	0.543773	0.225010	0.078484	00:15
2	0.332864	0.194363	0.070365	00:14
3	0.242706	0.188766	0.067659	00:14

In [11]:

learn.save('stage-1')

In [12]:

learn.unfreeze()
learn.fit_one_cycle(2, max_lr=slice(1e-6,1e-4))

epoch	train_loss	valid_loss	error_rate	time
0	0.231939	0.180866	0.062246	00:18
1	0.207379	0.175878	0.062246	00:17

In [13]:

interp = ClassificationInterpretation.from_learner(learn)
interp.most_confused(min_val=2)

Out[13]:

[('Egyptian_Mau', 'Bengal', 8),
 ('Ragdoll', 'Birman', 6),
 ('american_pit_bull_terrier', 'staffordshire_bull_terrier', 6),
 ('beagle', 'basset_hound', 5),
 ('american_bulldog', 'staffordshire_bull_terrier', 4),
 ('american_bulldog', 'american_pit_bull_terrier', 3),
 ('chihuahua', 'miniature_pinscher', 3),
 ('english_setter', 'english_cocker_spaniel', 3),
 ('samoyed', 'great_pyrenees', 3),
 ('Russian_Blue', 'British_Shorthair', 2),
 ('Siamese', 'Birman', 2),
 ('american_pit_bull_terrier', 'american_bulldog', 2),
 ('chihuahua', 'staffordshire_bull_terrier', 2),
 ('saint_bernard', 'american_bulldog', 2),
 ('yorkshire_terrier', 'havanese', 2)]

In [ ]: