Image-Retrieval/student.py at main · Riya-11/Image-Retrieval · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
from keras.utils import np_utils, to_categorical
from keras.models import load_model,Model
from keras.datasets import cifar10
import numpy as np
import matplotlib.pyplot as plt
from keras.callbacks import ModelCheckpoint
from keras.applications.resnet50 import ResNet50,preprocess_input
import keras
from sklearn.model_selection import train_test_split
from keras.activations import sigmoid,softmax
from keras.layers import Dense, Flatten, Activation, Conv2D, LeakyReLU, MaxPooling2D
from keras import models,layers
from keras.optimizers import SGD,Adam,RMSprop
from keras import backend as K
from keras import Sequential, Input
import matplotlib.image as mpimg
import tensorflow as tf
tf.config.run_functions_eagerly(True)
# %matplotlib inline
from distiller import Distiller

def get_data():
    (train_images, train_labels), (test_images, test_labels) = cifar10.load_data()
    train_images, test_images = train_images / 255.0, test_images / 255.0

    return train_images,train_labels,test_images,test_labels

def get_student_model():
    model = Sequential()
    model.add(Conv2D(32, (3, 3), activation='relu', input_shape=(32, 32, 3)))
    model.add(Conv2D(64, (3, 3), activation='relu'))#new

    model.add(MaxPooling2D((2, 2)))

    model.add(Conv2D(64, (3, 3), activation='relu'))
    model.add(Conv2D(64, (3, 3), activation='relu'))#new

    model.add(MaxPooling2D((2, 2)))

    model.add(Conv2D(64, (3, 3), activation='relu')) #new

    model.add(Flatten())

    model.add(Dense(128, activation='relu')) #new
    model.add(Dense(64, activation='relu'))
    model.add(Dense(32, activation='relu'))
    model.add(Dense(10))

    return model

class CustomCallback(keras.callbacks.Callback):
    def on_train_begin(self, logs={}):
        self.prev_acc = -9
        self.prev_val_loss = 100
        self.prev_val_acc = -9
        self.distillation_loss = []
        self.student_loss = []
        self.categorical_accuracy = []

    def on_train_end(self, logs=None):
        plt.style.use('seaborn')
        plt.plot(self.categorical_accuracy)
        plt.title('student model accuracy')
        # plt.ylabel('categorical_accuracy')
        plt.ylabel('accuracy')

        plt.xlabel('epoch')
        plt.legend(['train'], loc='upper left')
        plt.savefig('student_acc_adagrad.png')

        plt.show()
        # summarize history for loss
        plt.plot(self.student_loss)
        plt.plot(self.distillation_loss)
        plt.title('student model loss')
        plt.ylabel('loss')
        plt.xlabel('epoch')
        plt.legend(['train'], loc='upper left')
        plt.savefig('student_loss_adagrad.png')
        plt.show()

    def on_epoch_end(self, epoch, logs=None):
        keys = list(logs.keys())

        # curr_acc = logs['categorical_accuracy']
        curr_acc = logs['accuracy']
        curr_val_loss = logs['val_student_loss']
        curr_val_acc = logs['val_accuracy']
        self.student_loss.append(logs['student_loss'])
        self.distillation_loss.append(logs['distillation_loss'])
        self.categorical_accuracy.append(curr_acc)

        if curr_val_acc > self.prev_val_acc:
        # if curr_val_loss < self.prev_val_loss:
            # self.prev_acc = curr_acc
            self.prev_val_acc = curr_val_acc
            filename='new_arch/student-adagrad-tensorflow-ultra2-val-acc-weights-{}-{}.h5'.format(epoch,curr_val_acc)
            # distiller.save_weights(filename) #doesnt work!
            distiller.save(filename)

            print('\nModel saved as {} !'.format(filename))

        if epoch%20==0:clear_output()

train_images,train_labels,test_images,test_labels = get_data()

teacher_model = load_model('../teacher_3/weird_teacher_categorical-23-0.87.h5')#teacher_model_path here
teacher_distill = Model(inputs=teacher_model.input, outputs=teacher_model.layers[-2].output)

student = get_student_model()
student.compile(optimizer=keras.optimizers.Adam(0.001),
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

epochs = 100
distiller = Distiller(student=student, teacher=teacher_distill)

distiller.compile(
      optimizer=keras.optimizers.Adam(0.001),
      student_loss_fn=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
      metrics=['accuracy'],
      distillation_loss_fn=keras.losses.MeanSquaredError()
)

callbacks_list = [CustomCallback()]
distill_history = distiller.fit(train_images, train_labels, epochs=epochs,callbacks=callbacks_list,
                    validation_data=(test_images, test_labels))