Tensorflow Multiclass Image Classification Using Image Data Generator

 1|  import numpy as np
 2|  import pandas as pd
 3|  import tensorflow as tf
 4|  tf.random.set_seed(101)
 5|  from tensorflow.keras.models import Sequential
 6|  from tensorflow.keras.layers import Dense, Flatten, Dropout, MaxPooling2D
 7|  from tensorflow import keras
 8|  from tensorflow.keras import layers
 9|  from keras_preprocessing.image import ImageDataGenerator
10|  from efficientnet.tfkeras import EfficientNetB0
11|  from sklearn.metrics import accuracy_score 
12|  
13|  CLASSES = ['0', '1', '2', '3', '4']
14|  EPOCHS = 25
15|  
16|  IMG_DIM = 255
17|  BATCH_SIZE = 64
18|  
19|  #UNCOMMENT AND EDIT FOR ADDITIONAL AUGMENTATIONS
20|  train_datagen = ImageDataGenerator(
21|          rescale=1./255,
22|          #horizontal_flip=True,
23|          #vertical_flip=True,
24|          #featurewise_std_normalization=True,
25|          #rotation_range=180,
26|          #shear_range=2.0,
27|          #brightness_range=(0.0, 0.2)
28|          #zoom_range=(0,0.25,
29|          #channel_shift_range=30,
30|          #width_shift_range=0.2,
31|          #height_shift_range=0.2
32|          )
33|  
34|  test_datagen = ImageDataGenerator(
35|          rescale=1./255)
36|  
37|  def build_model():
38|      base_model = EfficientNetB0(input_shape=(IMG_DIM , IMG_DIM ,3),
39|                                               include_top=False,
40|                                               weights='imagenet')
41|      
42|      base_model.trainable = False
43|      model = Sequential(
44|          [base_model,
45|           tf.keras.layers.GlobalAveragePooling2D(),
46|          layers.Flatten(),
47|          Dense(512, activation= 'relu'), 
48|          Dropout(0.25), 
49|          Dense(256, activation='relu'), 
50|          Dropout(0.25), 
51|          Dense(128, activation='relu'), 
52|          Dropout(0.25),
53|          Dense(len(CLASSES), activation='softmax')]
54|          )
55|      return model
56|  
57|  train = pd.read_csv('data/train.csv')
58|  val = pd.read_csv('data/validation.csv')
59|  test = pd.read_csv('data/test.csv')
60|  train = train.sample(frac=1).reset_index(drop=True)
61|  val = val.sample(frac=1).reset_index(drop=True)
62|  
63|  train['label'] = train['label'].astype(str)
64|  val['label'] = val['label'].astype(str)
65|  test['label'] = test['label'].astype(str)
66|  
67|  train_generator = train_datagen.flow_from_dataframe(train, 
68|                                                      directory='data/images', 
69|                                                      x_col='image_id', #IMAGE FILENAME IN TRAIN DF
70|                                                      y_col='label', #IMAGE TARGET IN TRAIN DF
71|                                                      target_size=(IMG_DIM , IMG_DIM ),
72|                                                      batch_size=BATCH_SIZE ,
73|                                                      class_mode='sparse')
74|  val_generator = test_datagen.flow_from_dataframe(val, 
75|                                                  directory='data/images', 
76|                                                  x_col='image_id', #IMAGE FILENAME IN VAL DF
77|                                                  y_col='label', #IMAGE TARGET IN VAL DF
78|                                                  shuffle=False,
79|                                                  target_size=(IMG_DIM ,IMG_DIM ),
80|                                                  batch_size=BATCH_SIZE ,
81|                                                  class_mode='sparse')
82|  
83|  test_generator = test_datagen.flow_from_dataframe(test, 
84|                                                  directory='data/images', 
85|                                                  x_col='image_id', #IMAGE FILENAME IN TEST DF
86|                                                  y_col='label', #IMAGE TARGET IN TEST DF
87|                                                  shuffle=False,
88|                                                  target_size=(IMG_DIM ,IMG_DIM ),
89|                                                  batch_size=BATCH_SIZE ,
90|                                                  class_mode='sparse')
91|  
92|  tf.keras.backend.clear_session()
93|  
94|  model = build_model()
95|  
96|  STEP_SIZE_TRAIN=train_generator.n//train_generator.batch_size
97|  STEP_SIZE_VALID=val_generator.n//val_generator.batch_size
98|  
99|  model.compile(optimizer = tf.keras.optimizers.Adam(0.001),
100|                            loss='sparse_categorical_crossentropy',
101|                            metrics=['sparse_categorical_accuracy'])
102|  early_stop = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=10)
103|  history = model.fit(train_generator,
104|                      steps_per_epoch=STEP_SIZE_TRAIN,
105|                      validation_data=val_generator,
106|                      validation_steps=STEP_SIZE_VALID,
107|                      callbacks=[early_stop],
108|                      epochs=EPOCHS)
109|  
110|  
111|  #MAKE PREDICTIONS FOR VALIDATION SET AND CALCULATE ACCURACY
112|  predictions = model.predict(test_generator,
113|                       steps=None,
114|                       verbose=1)
115|  predictions = np.argmax(predictions, axis=-1)
116|  val['label'] = val['label'].astype(int)
117|  score = accuracy_score(val['label'], predictions)
118|  print('Accuracy:',score)