In [1]:

    

import numpy as np
import pandas as pd
import os

from keras.models import Sequential
from keras.layers.convolutional import Convolution2D
from keras.layers.pooling import MaxPooling2D
from keras.layers.core import Flatten, Dense, Activation, Dropout
from keras.preprocessing import image
from keras.callbacks import EarlyStopping, ModelCheckpoint
from keras.utils.np_utils import to_categorical
from sklearn.model_selection import train_test_split
from matplotlib import pyplot as plt


    

    




Using TensorFlow backend.

In [2]:

    

# My LeNet architecture
model = Sequential()

# conv filters of 5x5 each

# Layer 1
model.add(Convolution2D(32, 5, 5, input_shape=(192, 192, 3)))
model.add(Activation("relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))

# Layer 2
model.add(Convolution2D(64, 5, 5))
model.add(Activation("relu"))
model.add(MaxPooling2D(pool_size=(2, 2)))    

model.add(Flatten())

# Layer 3
model.add(Dense(1024))
model.add(Activation("relu"))
model.add(Dropout(0.5))

# Layer 4
model.add(Dense(512))
model.add(Activation("relu"))
model.add(Dropout(0.5))

# Layer 5
model.add(Dense(2))
model.add(Activation("softmax"))


    

In [3]:

    

model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])


    

In [69]:

    

X = []
y = []


    

In [5]:

    

with open("X.pkl", 'rb') as picklefile: 
    X = pickle.load(picklefile)


    

In [6]:

    

with open("y.pkl", 'rb') as picklefile: 
    y = pickle.load(picklefile)


    

In [70]:

    

# set folder path
folderpath = 'Images/Train/Undistorted/'

# load image arrays
for filename in os.listdir(folderpath):
    if filename != '.DS_Store':
        imagepath = folderpath + filename
        img = image.load_img(imagepath, target_size=(192,192))
        X.append(np.asarray(img))
        y.append(0)
    else:
        print filename, 'not a pic'


    

In [4]:

    

import pickle


    

In [12]:

    

with open('undistorted_X.pkl', 'wb') as picklefile:
    pickle.dump(X, picklefile)


    

In [13]:

    

with open('undistorted_y.pkl', 'wb') as picklefile:
    pickle.dump(y, picklefile)


    

In [7]:

    

# set folder path
folderpath = 'Images/Train/DigitalBlur2/'

# load image arrays
for filename in os.listdir(folderpath):
    if filename != '.DS_Store':
        imagepath = folderpath + filename
        img = image.load_img(imagepath, target_size=(192,192))
        X.append(np.asarray(img))
        y.append(1)
    else:
        print filename, 'not a pic'


    

In [9]:

    

len(y)


    

    
Out[9]:





2731

In [17]:

    

with open('X.pkl', 'wb') as picklefile:
    pickle.dump(X, picklefile)


    

In [18]:

    

with open('y.pkl', 'wb') as picklefile:
    pickle.dump(y, picklefile)


    

In [10]:

    

X_stacked = np.stack(X)
X_norm = X_stacked/255.
y_cat = to_categorical(y)


    

In [11]:

    

X_train, X_test, y_train, y_test = train_test_split(X_norm, y_cat, train_size=2500, random_state=42)


    

In [12]:

    

# Data augmenter
dg = image.ImageDataGenerator(horizontal_flip=True, vertical_flip=True)


    

In [13]:

    

model.load_weights('lenet_weights.h5')


    

In [14]:

    

cb_es = EarlyStopping(monitor='val_acc', patience=2, verbose=1)
cb_mc = ModelCheckpoint('lenet_weights2.h5', monitor='val_acc', verbose=1, save_best_only=True, save_weights_only=True)


    

In [15]:

    

# Fit generator
# play with samples/epoch, nb_epoch, val_samples.

model.fit_generator(dg.flow(X_train, y_train), samples_per_epoch=3000, nb_epoch=30, validation_data=dg.flow(X_test, y_test), nb_val_samples=300, callbacks=[cb_es, cb_mc])


    

    




Epoch 1/30
2980/3000 [============================>.] - ETA: 2s - loss: 0.1901 - acc: 0.9460  





    




//anaconda/lib/python2.7/site-packages/keras/engine/training.py:1573: UserWarning: Epoch comprised more than `samples_per_epoch` samples, which might affect learning results. Set `samples_per_epoch` correctly to avoid this warning.
  warnings.warn('Epoch comprised more than '






    




Epoch 00000: val_acc improved from -inf to 0.96636, saving model to lenet_weights2.h5
3012/3000 [==============================] - 416s - loss: 0.1883 - acc: 0.9465 - val_loss: 0.1004 - val_acc: 0.9664
Epoch 2/30
2980/3000 [============================>.] - ETA: 2s - loss: 0.1020 - acc: 0.9708  Epoch 00001: val_acc improved from 0.96636 to 0.99669, saving model to lenet_weights2.h5
3012/3000 [==============================] - 386s - loss: 0.1013 - acc: 0.9708 - val_loss: 0.0224 - val_acc: 0.9967
Epoch 3/30
2980/3000 [============================>.] - ETA: 2s - loss: 0.0736 - acc: 0.9809  Epoch 00002: val_acc did not improve
3012/3000 [==============================] - 380s - loss: 0.0729 - acc: 0.9811 - val_loss: 0.0902 - val_acc: 0.9664
Epoch 4/30
2980/3000 [============================>.] - ETA: 2s - loss: 0.1115 - acc: 0.9718  Epoch 00003: val_acc did not improve
3012/3000 [==============================] - 377s - loss: 0.1118 - acc: 0.9718 - val_loss: 0.0128 - val_acc: 0.9967
Epoch 5/30
2984/3000 [============================>.] - ETA: 1s - loss: 0.1042 - acc: 0.9715  Epoch 00004: val_acc did not improve
3016/3000 [==============================] - 379s - loss: 0.1051 - acc: 0.9715 - val_loss: 0.1197 - val_acc: 0.9694
Epoch 00004: early stopping






    
Out[15]:





<keras.callbacks.History at 0x25f2c9ad0>

In [ ]:

    




    

In [32]:

    

clean_pics = []
blurry_pics = []


    

In [6]:

    

backBlur_pics = []


    

In [33]:

    

# set folder path
folderpath = 'Images/clearSamples/'

# load image arrays
for filename in os.listdir(folderpath):
    if filename != '.DS_Store':
        imagepath = folderpath + filename
        img = image.load_img(imagepath, target_size=(192,192))
        clean_pics.append(np.asarray(img))
    else:
        print filename, 'not a pic'


    

In [34]:

    

# set folder path
folderpath = 'Images/natblurSamples/'

# load image arrays
for filename in os.listdir(folderpath):
    if filename != '.DS_Store':
        imagepath = folderpath + filename
        img = image.load_img(imagepath, target_size=(192,192))
        blurry_pics.append(np.asarray(img))
    else:
        print filename, 'not a pic'


    

In [7]:

    

# set folder path
folderpath = 'Images/backBlurSamples/'

# load image arrays
for filename in os.listdir(folderpath):
    if filename != '.DS_Store':
        imagepath = folderpath + filename
        img = image.load_img(imagepath, target_size=(192,192))
        backBlur_pics.append(np.asarray(img))
    else:
        print filename, 'not a pic'


    

In [82]:

    

len(backBlur_pics)


    

    
Out[82]:





7

In [9]:

    

backBlur_pics_array = np.stack(backBlur_pics)/255.


    

In [36]:

    

len(blurry_pics)


    

    
Out[36]:





28

In [37]:

    

clean_pics_array = np.stack(clean_pics)/255.


    

In [38]:

    

blurry_pics_array = np.stack(blurry_pics)/255.


    

In [42]:

    

blurry_pics_array.shape


    

    
Out[42]:





(28, 192, 192, 3)

In [44]:

    

model.predict_classes(clean_pics_array)


    

    




16/16 [==============================] - 0s






    
Out[44]:





array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0])

In [91]:

    

model.predict_proba(clean_pics_array)


    

    




16/16 [==============================] - 0s






    
Out[91]:





array([[  9.99999166e-01,   8.47247861e-07],
       [  9.76107121e-01,   2.38929167e-02],
       [  9.93018031e-01,   6.98202476e-03],
       [  9.99996185e-01,   3.77086599e-06],
       [  1.00000000e+00,   4.99104610e-14],
       [  1.00000000e+00,   8.84988229e-16],
       [  9.58419383e-01,   4.15805727e-02],
       [  1.00000000e+00,   0.00000000e+00],
       [  1.00000000e+00,   2.15501283e-09],
       [  9.83334243e-01,   1.66657493e-02],
       [  1.00000000e+00,   1.06239604e-17],
       [  9.99967575e-01,   3.24022658e-05],
       [  1.00000000e+00,   3.08845935e-11],
       [  1.00000000e+00,   2.12371787e-11],
       [  1.00000000e+00,   7.84776886e-17],
       [  1.00000000e+00,   3.81958605e-21]], dtype=float32)

In [45]:

    

model.predict_classes(blurry_pics_array)


    

    




28/28 [==============================] - 0s






    
Out[45]:





array([0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 1, 1, 0, 1, 1,
       0, 1, 1, 0, 1])

In [90]:

    

model.predict_proba(blurry_pics_array)


    

    




28/28 [==============================] - 0s






    
Out[90]:





array([[  9.99991655e-01,   8.28602788e-06],
       [  2.93354075e-02,   9.70664620e-01],
       [  9.51888204e-01,   4.81118150e-02],
       [  9.91434097e-01,   8.56589619e-03],
       [  2.93354075e-02,   9.70664620e-01],
       [  9.99984622e-01,   1.53655328e-05],
       [  9.97849107e-01,   2.15089903e-03],
       [  1.00000000e+00,   4.09413238e-20],
       [  9.99999404e-01,   5.87542445e-07],
       [  8.63840759e-01,   1.36159256e-01],
       [  7.04566777e-01,   2.95433223e-01],
       [  4.54591438e-02,   9.54540849e-01],
       [  2.93992683e-02,   9.70600724e-01],
       [  6.12833835e-02,   9.38716590e-01],
       [  2.93354075e-02,   9.70664620e-01],
       [  9.92878497e-01,   7.12153688e-03],
       [  1.00000000e+00,   2.55035673e-08],
       [  6.80495024e-01,   3.19504976e-01],
       [  2.97666360e-02,   9.70233440e-01],
       [  5.34660518e-02,   9.46533978e-01],
       [  6.11283958e-01,   3.88716072e-01],
       [  3.47494967e-02,   9.65250492e-01],
       [  5.42964637e-02,   9.45703566e-01],
       [  6.72513604e-01,   3.27486426e-01],
       [  8.49336684e-02,   9.15066361e-01],
       [  3.47424522e-02,   9.65257525e-01],
       [  9.25547361e-01,   7.44526684e-02],
       [  2.93354075e-02,   9.70664620e-01]], dtype=float32)

In [10]:

    

model.predict_proba(backBlur_pics_array)


    

    




7/7 [==============================] - 0s






    
Out[10]:





array([[  1.00000000e+00,   7.19240607e-31],
       [  9.99990940e-01,   9.08570655e-06],
       [  1.00000000e+00,   1.16131817e-14],
       [  1.00000000e+00,   3.99548874e-08],
       [  1.00000000e+00,   3.96198019e-10],
       [  1.00000000e+00,   1.00476560e-09],
       [  9.99983668e-01,   1.63695859e-05]], dtype=float32)

In [89]:

    

plt.imshow(blurry_pics_array[0])
plt.show()


    

In [93]:

    

model.save('lenet_3rdPass.h5')


    

In [86]:

    

model.save_weights('test_weights.h5')


    

In [20]:

    

backBlur_pics = []
backBlur_filenames = []

# set folder path
folderpath = 'Images/backBlurAll_longIter/'

# load image arrays
for filename in os.listdir(folderpath):
    if filename != '.DS_Store':
        backBlur_filenames.append(filename)
        imagepath = folderpath + filename
        img = image.load_img(imagepath, target_size=(192,192))
        backBlur_pics.append(np.asarray(img))
    else:
        print filename, 'not a pic'

df_backBlur = pd.DataFrame(backBlur_filenames, columns=['filename'])


    

    




.DS_Store not a pic

In [21]:

    

backBlur_pics_array = np.stack(backBlur_pics)/255.


    

In [22]:

    

df_backBlur['blur_class'] = model.predict_classes(backBlur_pics_array)


    

    




1000/1000 [==============================] - 33s     

In [23]:

    

if not os.path.exists(folderpath+'blurry'):
    os.mkdir(folderpath+'blurry')

for index, row in df_backBlur.iterrows():
    if row['blur_class'] == 1:
        oldpath = folderpath + row['filename']
        newpath = folderpath + 'blurry/' + row['filename']
        os.rename(oldpath, newpath)


    

In [16]:

    

natBlur_pics = []
natBlur_filenames = []

# set folder path
folderpath = 'Images/natBlurAll_longIter/'

# load image arrays
for filename in os.listdir(folderpath):
    if filename != '.DS_Store':
        natBlur_filenames.append(filename)
        imagepath = folderpath + filename
        img = image.load_img(imagepath, target_size=(192,192))
        natBlur_pics.append(np.asarray(img))
    else:
        print filename, 'not a pic'

df_natBlur = pd.DataFrame(natBlur_filenames, columns=['filename'])


    

    




.DS_Store not a pic

In [17]:

    

natBlur_pics_array = np.stack(natBlur_pics)/255.


    

In [18]:

    

df_natBlur['blur_class'] = model.predict_classes(natBlur_pics_array)


    

    




411/411 [==============================] - 13s     

In [19]:

    

if not os.path.exists(folderpath+'blurry'):
    os.mkdir(folderpath+'blurry')

for index, row in df_natBlur.iterrows():
    if row['blur_class'] == 1:
        oldpath = folderpath + row['filename']
        newpath = folderpath + 'blurry/' + row['filename']
        os.rename(oldpath, newpath)


    

In [24]:

    

m3Blur_pics = []
m3Blur_filenames = []

# set folder path
folderpath = 'Images/M3Samples/'

# load image arrays
for filename in os.listdir(folderpath):
    if filename != '.DS_Store':
        m3Blur_filenames.append(filename)
        imagepath = folderpath + filename
        img = image.load_img(imagepath, target_size=(192,192))
        m3Blur_pics.append(np.asarray(img))
    else:
        print filename, 'not a pic'

df_m3Blur = pd.DataFrame(m3Blur_filenames, columns=['filename'])


    

In [25]:

    

m3Blur_pics_array = np.stack(m3Blur_pics)/255.


    

In [26]:

    

df_m3Blur['blur_class'] = model.predict_classes(m3Blur_pics_array)


    

    




4/4 [==============================] - 0s

In [27]:

    

if not os.path.exists(folderpath+'blurry'):
    os.mkdir(folderpath+'blurry')

for index, row in df_m3Blur.iterrows():
    if row['blur_class'] == 1:
        oldpath = folderpath + row['filename']
        newpath = folderpath + 'blurry/' + row['filename']
        os.rename(oldpath, newpath)


    

In [28]:

    

df_m3Blur


    

    
Out[28]:






  

    

      

      
filename
      
blur_class
    
  
  

    

      
0
      
225588_10100452581844163_5171555_n.jpg
      
0
    
    

      
1
      
225923_10100452581864123_5207008_n.jpg
      
0
    
    

      
2
      
230393_10100452581789273_2903274_n.jpg
      
0
    
    

      
3
      
231133_10100452581739373_3369180_n.jpg
      
0
    
  

In [29]:

    

clearSample_pics = []
clearSample_filenames = []

# set folder path
folderpath = 'Images/clearSamples/'

# load image arrays
for filename in os.listdir(folderpath):
    if filename != '.DS_Store':
        clearSample_filenames.append(filename)
        imagepath = folderpath + filename
        img = image.load_img(imagepath, target_size=(192,192))
        clearSample_pics.append(np.asarray(img))
    else:
        print filename, 'not a pic'

df_clearSample = pd.DataFrame(clearSample_filenames, columns=['filename'])
clearSample_pics_array = np.stack(clearSample_pics)/255.
df_clearSample['blur_class'] = model.predict_classes(clearSample_pics_array)


    

    




16/16 [==============================] - 0s

In [30]:

    

if not os.path.exists(folderpath+'blurry'):
    os.mkdir(folderpath+'blurry')

for index, row in df_clearSample.iterrows():
    if row['blur_class'] == 1:
        oldpath = folderpath + row['filename']
        newpath = folderpath + 'blurry/' + row['filename']
        os.rename(oldpath, newpath)


    

In [31]:

    

blurSample_pics = []
blurSample_filenames = []

# set folder path
folderpath = 'Images/natBlurSamples/'

# load image arrays
for filename in os.listdir(folderpath):
    if filename != '.DS_Store':
        blurSample_filenames.append(filename)
        imagepath = folderpath + filename
        img = image.load_img(imagepath, target_size=(192,192))
        blurSample_pics.append(np.asarray(img))
    else:
        print filename, 'not a pic'

df_blurSample = pd.DataFrame(blurSample_filenames, columns=['filename'])
blurSample_pics_array = np.stack(blurSample_pics)/255.
df_blurSample['blur_class'] = model.predict_classes(blurSample_pics_array)


    

    




.DS_Store not a pic
25/25 [==============================] - 0s

In [33]:

    

if not os.path.exists(folderpath+'blurry'):
    os.mkdir(folderpath+'blurry')

for index, row in df_blurSample.iterrows():
    if row['blur_class'] == 1:
        oldpath = folderpath + row['filename']
        newpath = folderpath + 'blurry/' + row['filename']
        os.rename(oldpath, newpath)


    

In [ ]:

    




    

In [5]:

    

img = image.load_img('Images/NaturalBlurSet.xlsx', target_size=(192,192))


    

    




---------------------------------------------------------------------------
IOError                                   Traceback (most recent call last)
<ipython-input-5-6a4ba8227f17> in <module>()
----> 1 img = image.load_img('Images/NaturalBlurSet.xlsx', target_size=(192,192))

//anaconda/lib/python2.7/site-packages/keras/preprocessing/image.pyc in load_img(path, grayscale, target_size)
    294         raise ImportError('Could not import PIL.Image. '
    295                           'The use of `array_to_img` requires PIL.')
--> 296     img = pil_image.open(path)
    297     if grayscale:
    298         img = img.convert('L')

//anaconda/lib/python2.7/site-packages/PIL/Image.pyc in open(fp, mode)
   2347 
   2348     raise IOError("cannot identify image file %r"
-> 2349                   % (filename if filename else fp))
   2350 
   2351 #

IOError: cannot identify image file 'Images/NaturalBlurSet.xlsx'

In [8]:

    

'.JPG'.lower()


    

    
Out[8]:





'.jpg'

In [ ]: