Join the “Zero to Data Analyst” Bootcamp. Limited seats.

A 20-week program covering 7 courses, 12 assignments, 4 real-world projects, and 12 months of career support.

v-snehith999
/
predicting-unclear-images-of-objects-neuralnetworks

Updated a year ago

Duplicate

Notebook Files Collaborators

Version 1 (a year ago)

Compare Present

CIFAR-10 Multiple Classes

Constructing a CNN using Keras and Tensorflow for training and predicting animal from a given unclear image. Here we are using CIFAR-10 Data set. CIFAR-10 dataset which consists of 10 different image types.

The Data

CIFAR-10 is a dataset of 50,000 32x32 color training images, labeled over 10 categories, and 10,000 test images.

In [ ]:

from keras.datasets import cifar10

(x_train, y_train), (x_test, y_test) = cifar10.load_data()

Downloading data from https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz

In [2]:

x_train.shape

Out[2]:

(50000, 32, 32, 3)

In [3]:

x_train[0].shape

Out[3]:

(32, 32, 3)

In [ ]:

import matplotlib.pyplot as plt

In [5]:

# FROG
plt.imshow(x_train[0])

Out[5]:

<matplotlib.image.AxesImage at 0x295e6e1cba8>

In [6]:

# HORSE
plt.imshow(x_train[12])

Out[6]:

<matplotlib.image.AxesImage at 0x295e6ebb0b8>

PreProcessing

In [7]:

x_train[0]

Out[7]:

array([[[ 59,  62,  63],
        [ 43,  46,  45],
        [ 50,  48,  43],
        ...,
        [158, 132, 108],
        [152, 125, 102],
        [148, 124, 103]],

       [[ 16,  20,  20],
        [  0,   0,   0],
        [ 18,   8,   0],
        ...,
        [123,  88,  55],
        [119,  83,  50],
        [122,  87,  57]],

       [[ 25,  24,  21],
        [ 16,   7,   0],
        [ 49,  27,   8],
        ...,
        [118,  84,  50],
        [120,  84,  50],
        [109,  73,  42]],

       ...,

       [[208, 170,  96],
        [201, 153,  34],
        [198, 161,  26],
        ...,
        [160, 133,  70],
        [ 56,  31,   7],
        [ 53,  34,  20]],

       [[180, 139,  96],
        [173, 123,  42],
        [186, 144,  30],
        ...,
        [184, 148,  94],
        [ 97,  62,  34],
        [ 83,  53,  34]],

       [[177, 144, 116],
        [168, 129,  94],
        [179, 142,  87],
        ...,
        [216, 184, 140],
        [151, 118,  84],
        [123,  92,  72]]], dtype=uint8)

In [8]:

x_train[0].shape

Out[8]:

(32, 32, 3)

In [9]:

x_train.max()

Out[9]:

In [10]:

x_train = x_train/225

In [11]:

x_test = x_test/255

In [12]:

x_train.shape

Out[12]:

(50000, 32, 32, 3)

In [13]:

x_test.shape

Out[13]:

(10000, 32, 32, 3)

Labels

In [14]:

from keras.utils import to_categorical

In [15]:

y_train.shape

Out[15]:

(50000, 1)

In [16]:

y_train[0]

Out[16]:

array([6], dtype=uint8)

In [17]:

y_cat_train = to_categorical(y_train,10)

In [18]:

y_cat_train.shape

Out[18]:

(50000, 10)

In [19]:

y_cat_train[0]

Out[19]:

array([0., 0., 0., 0., 0., 0., 1., 0., 0., 0.], dtype=float32)

In [20]:

y_cat_test = to_categorical(y_test,10)

Building the Model

In [21]:

from keras.models import Sequential
from keras.layers import Dense, Conv2D, MaxPool2D, Flatten

In [22]:

model = Sequential()

## FIRST SET OF LAYERS

# CONVOLUTIONAL LAYER
model.add(Conv2D(filters=32, kernel_size=(4,4),input_shape=(32, 32, 3), activation='relu',))
# POOLING LAYER
model.add(MaxPool2D(pool_size=(2, 2)))

## SECOND SET OF LAYERS

# CONVOLUTIONAL LAYER
model.add(Conv2D(filters=32, kernel_size=(4,4),input_shape=(32, 32, 3), activation='relu',))
# POOLING LAYER
model.add(MaxPool2D(pool_size=(2, 2)))

# FLATTEN IMAGES FROM 28 by 28 to 764 BEFORE FINAL LAYER
model.add(Flatten())

# 256 NEURONS IN DENSE HIDDEN LAYER (YOU CAN CHANGE THIS NUMBER OF NEURONS)
model.add(Dense(256, activation='relu'))

# LAST LAYER IS THE CLASSIFIER, THUS 10 POSSIBLE CLASSES
model.add(Dense(10, activation='softmax'))


model.compile(loss='categorical_crossentropy',
              optimizer='rmsprop',
              metrics=['accuracy'])

In [23]:

model.summary()

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 29, 29, 32)        1568      
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 14, 14, 32)        0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 11, 11, 32)        16416     
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 5, 5, 32)          0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 800)               0         
_________________________________________________________________
dense_1 (Dense)              (None, 256)               205056    
_________________________________________________________________
dense_2 (Dense)              (None, 10)                2570      
=================================================================
Total params: 225,610
Trainable params: 225,610
Non-trainable params: 0
_________________________________________________________________

In [25]:

model.fit(x_train,y_cat_train,verbose=1,epochs=10)

Epoch 1/10
50000/50000 [==============================] - 5s 109us/step - loss: 0.8937 - acc: 0.6919
Epoch 2/10
50000/50000 [==============================] - 5s 108us/step - loss: 0.8094 - acc: 0.7249
Epoch 3/10
50000/50000 [==============================] - 5s 109us/step - loss: 0.7420 - acc: 0.7461
Epoch 4/10
50000/50000 [==============================] - 5s 109us/step - loss: 0.6902 - acc: 0.7656
Epoch 5/10
50000/50000 [==============================] - 5s 109us/step - loss: 0.6466 - acc: 0.7780
Epoch 6/10
50000/50000 [==============================] - 5s 110us/step - loss: 0.6105 - acc: 0.7958
Epoch 7/10
50000/50000 [==============================] - 5s 110us/step - loss: 0.5891 - acc: 0.8020
Epoch 8/10
50000/50000 [==============================] - 5s 109us/step - loss: 0.5631 - acc: 0.8124
Epoch 9/10
50000/50000 [==============================] - 6s 110us/step - loss: 0.5416 - acc: 0.8203
Epoch 10/10
50000/50000 [==============================] - 5s 109us/step - loss: 0.5252 - acc: 0.8275

Out[25]:

<keras.callbacks.History at 0x295dce78dd8>

In [26]:

# Careful, don't overwrite our file!
# model.save('cifar_10epochs.h5')

In [27]:

model.metrics_names

Out[27]:

['loss', 'acc']

In [29]:

model.evaluate(x_test,y_cat_test)

10000/10000 [==============================] - 1s 56us/step

Out[29]:

[1.3332478387832642, 0.6444]

In [30]:

from sklearn.metrics import classification_report

predictions = model.predict_classes(x_test)

In [31]:

print(classification_report(y_test,predictions))

             precision    recall  f1-score   support

          0       0.80      0.59      0.68      1000
          1       0.83      0.76      0.80      1000
          2       0.44      0.65      0.52      1000
          3       0.50      0.40      0.44      1000
          4       0.50      0.75      0.60      1000
          5       0.52      0.57      0.54      1000
          6       0.70      0.74      0.72      1000
          7       0.88      0.56      0.69      1000
          8       0.82      0.74      0.78      1000
          9       0.81      0.68      0.74      1000

avg / total       0.68      0.64      0.65     10000

Optional: Large Model

In [33]:

model = Sequential()

## FIRST SET OF LAYERS

# CONVOLUTIONAL LAYER
model.add(Conv2D(filters=32, kernel_size=(4,4),input_shape=(32, 32, 3), activation='relu',))
# CONVOLUTIONAL LAYER
model.add(Conv2D(filters=32, kernel_size=(4,4),input_shape=(32, 32, 3), activation='relu',))

# POOLING LAYER
model.add(MaxPool2D(pool_size=(2, 2)))

## SECOND SET OF LAYERS

# CONVOLUTIONAL LAYER
model.add(Conv2D(filters=64, kernel_size=(4,4),input_shape=(32, 32, 3), activation='relu',))
# CONVOLUTIONAL LAYER
model.add(Conv2D(filters=64, kernel_size=(4,4),input_shape=(32, 32, 3), activation='relu',))

# POOLING LAYER
model.add(MaxPool2D(pool_size=(2, 2)))

# FLATTEN IMAGES FROM 28 by 28 to 764 BEFORE FINAL LAYER
model.add(Flatten())

# 512 NEURONS IN DENSE HIDDEN LAYER (YOU CAN CHANGE THIS NUMBER OF NEURONS)
model.add(Dense(512, activation='relu'))

# LAST LAYER IS THE CLASSIFIER, THUS 10 POSSIBLE CLASSES
model.add(Dense(10, activation='softmax'))


model.compile(loss='categorical_crossentropy',
              optimizer='rmsprop',
              metrics=['accuracy'])

In [34]:

model.fit(x_train,y_cat_train,verbose=1,epochs=20)

Epoch 1/20
50000/50000 [==============================] - 7s 148us/step - loss: 1.6025 - acc: 0.4197
Epoch 2/20
50000/50000 [==============================] - 8s 159us/step - loss: 1.1793 - acc: 0.5849
Epoch 3/20
50000/50000 [==============================] - 8s 155us/step - loss: 1.0083 - acc: 0.6508
Epoch 4/20
50000/50000 [==============================] - 8s 156us/step - loss: 0.9479 - acc: 0.6782
Epoch 5/20
50000/50000 [==============================] - 8s 154us/step - loss: 0.9145 - acc: 0.6952
Epoch 6/20
50000/50000 [==============================] - 8s 151us/step - loss: 0.9005 - acc: 0.7015
Epoch 7/20
50000/50000 [==============================] - 8s 151us/step - loss: 0.8867 - acc: 0.7061
Epoch 8/20
50000/50000 [==============================] - 8s 152us/step - loss: 0.8796 - acc: 0.7121
Epoch 9/20
50000/50000 [==============================] - 8s 151us/step - loss: 0.8693 - acc: 0.7158
Epoch 10/20
50000/50000 [==============================] - 8s 151us/step - loss: 0.8591 - acc: 0.7194
Epoch 11/20
50000/50000 [==============================] - 7s 145us/step - loss: 0.8586 - acc: 0.7235
Epoch 12/20
50000/50000 [==============================] - 8s 151us/step - loss: 0.8497 - acc: 0.7248
Epoch 13/20
50000/50000 [==============================] - 7s 148us/step - loss: 0.8462 - acc: 0.7296
Epoch 14/20
50000/50000 [==============================] - 7s 139us/step - loss: 0.8422 - acc: 0.7263
Epoch 15/20
50000/50000 [==============================] - 7s 147us/step - loss: 0.8321 - acc: 0.7330
Epoch 16/20
50000/50000 [==============================] - 7s 143us/step - loss: 0.8154 - acc: 0.7363
Epoch 17/20
50000/50000 [==============================] - 7s 147us/step - loss: 0.8241 - acc: 0.7353
Epoch 18/20
50000/50000 [==============================] - 7s 138us/step - loss: 0.8061 - acc: 0.7411
Epoch 19/20
50000/50000 [==============================] - 7s 140us/step - loss: 0.8120 - acc: 0.7409
Epoch 20/20
50000/50000 [==============================] - 7s 142us/step - loss: 0.8097 - acc: 0.7437

Out[34]:

<keras.callbacks.History at 0x295e6eb3e80>

In [35]:

model.evaluate(x_test,y_cat_test)

10000/10000 [==============================] - 1s 75us/step

Out[35]:

[0.9843294318199157, 0.6902]

In [36]:

from sklearn.metrics import classification_report

predictions = model.predict_classes(x_test)

In [37]:

print(classification_report(y_test,predictions))

             precision    recall  f1-score   support

          0       0.82      0.61      0.70      1000
          1       0.77      0.88      0.82      1000
          2       0.64      0.57      0.60      1000
          3       0.66      0.30      0.41      1000
          4       0.66      0.67      0.66      1000
          5       0.45      0.81      0.58      1000
          6       0.80      0.75      0.77      1000
          7       0.76      0.69      0.72      1000
          8       0.78      0.81      0.80      1000
          9       0.75      0.82      0.79      1000

avg / total       0.71      0.69      0.69     10000

In [38]:

model.save('larger_CIFAR10_model.h5')

v-snehith999/predicting-unclear-images-of-objects-neuralnetworks