%reload_ext autoreload
%autoreload 2
%matplotlib inline

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

PATH = Path("../../data/lv_imgs/")

!ls {PATH}

models	train

!ls {PATH}/train

fake_bags  real_bags

!ls {PATH}/train/fake_bags | wc -l

51

!ls {PATH}/train/real_bags | wc -l

70

sample_img = !ls {PATH}/train/fake_bags | head -n 1
#sample_img
img = plt.imread(f'{PATH}/train/fake_bags/{sample_img[0]}')
plt.imshow(img)

<matplotlib.image.AxesImage at 0x7fe0308f64e0>

np.random.seed(561)
data = ImageDataBunch.from_folder(PATH, 
                                  ds_tfms=get_transforms(), 
                                  size=224,
                                  valid_pct= .25,
                                  bs=16
                                  ).normalize(imagenet_stats)

learn = create_cnn(data, 
                   models.resnet50, 
                   metrics=error_rate,
                   ps=.7)

learn.fit_one_cycle(4,
                    max_lr=slice(3e-05, 3e-04))

Total time: 00:12

epoch	train_loss	valid_loss	error_rate
1	1.106430	0.649481	0.366667
2	1.373133	0.652975	0.466667
3	1.250432	0.668808	0.400000
4	1.103542	0.672520	0.400000

interp = ClassificationInterpretation.from_learner(learn)
losses,idxs=interp.top_losses()
interp.plot_top_losses(9)

interp.plot_confusion_matrix()

learn.unfreeze()

learn.lr_find()
learn.recorder.plot()

LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.

learn.fit_one_cycle(2, 
                    max_lr=3e-4)

Total time: 00:07

epoch	train_loss	valid_loss	error_rate
1	0.910160	0.800928	0.400000
2	0.762213	0.756685	0.366667

interp = ClassificationInterpretation.from_learner(learn)
losses,idxs=interp.top_losses()

interp.plot_confusion_matrix()

interp.plot_top_losses(9)

from fastai.widgets import *

ds, idxs = DatasetFormatter().from_toplosses(learn, ds_type=DatasetType.Train)

ImageCleaner(ds, idxs, PATH)