trt-custom-plugin/tf_utils/model.py at 45532a116791cd6ffb5a607db3adccadbeabe035 · codesteller/trt-custom-plugin · 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
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
import tensorflow as tf
from numpy import pi
from keras.backend import sigmoid
from keras.models import Sequential, load_model
from keras.layers import Layer, InputLayer, Dense, Flatten, Activation, Conv2D, MaxPooling2D, LeakyReLU
from keras.callbacks import ModelCheckpoint, TensorBoard
from keras.optimizers import Adam
from keras.backend import get_session
from keras.utils.generic_utils import get_custom_objects
import os
from glob import glob


class Model:
    def __init__(self,
                 input_shape=(224, 224, 3),
                 num_classes=2,
                 checkpoint_path="./checkpoint",
                 batch_size=32,
                 epochs=10,
                 learning_rate=0.001):
        """
        input_shape - In HWC format
        """
        self.model = Sequential()
        self.input_shape = input_shape
        self.num_classes = num_classes
        self.checkpoint_path = checkpoint_path
        self.batch_size = batch_size
        self.learning_rate = learning_rate
        self.epochs = epochs
        get_custom_objects().update({'gelu_activation': Activation(self.gelu_activation)})

    def build_model(self):
        self.model.add(InputLayer(input_shape=self.input_shape))
        self.model.add(Conv2D(32, kernel_size=(3, 3), activation='linear',
                              input_shape=self.input_shape, padding='same'))
        self.model.add(Activation('relu'))
        self.model.add(MaxPooling2D((2, 2), padding='same'))
        self.model.add(Conv2D(64, (3, 3), activation='linear', padding='same'))
        self.model.add(Activation('relu'))
        self.model.add(MaxPooling2D(pool_size=(2, 2), padding='same'))
        self.model.add(
            Conv2D(128, (3, 3), activation='linear', padding='same'))
        self.model.add(Activation('relu'))
        self.model.add(MaxPooling2D(pool_size=(2, 2), padding='same'))
        self.model.add(Flatten())
        self.model.add(Dense(128, activation="linear"))
        self.model.add(Activation('gelu_activation', name='GeluActivation'))
        self.model.add(Dense(self.num_classes, activation='softmax'))

        opt = Adam(learning_rate=0.001, beta_1=0.9, beta_2=0.999, amsgrad=False)

        self.model.compile(optimizer=opt,
                           loss='binary_crossentropy',
                           metrics=['accuracy'])
        print("Model built and compiled successfully")

    @staticmethod
    def gelu_activation(_input, alpha=1):
        return 0.5 * _input * (alpha + tf.tanh(tf.sqrt(2 / pi) * (_input + 0.044715 * _input * _input * _input)))

    def train_model(self, train_data_gen, valid_data_gen):
        checkpoint_dir = os.path.dirname(self.checkpoint_path)
        if not os.path.exists(checkpoint_dir):
            os.makedirs(checkpoint_dir)
        tensorboard_dir = os.path.join(checkpoint_dir, 'tensorboard')
        if not os.path.exists(tensorboard_dir):
            os.makedirs(tensorboard_dir)

        # Create a callback that saves the model's weights
        cp_callback = ModelCheckpoint(filepath=self.checkpoint_path, save_weights_only=True, verbose=1, period=1)
        # cp_callback = ModelCheckpoint(filepath=self.checkpoint_path, monitor='val_acc', verbose=1,
        #                               save_best_only=True, mode='max')
        tb_callback = TensorBoard(log_dir=tensorboard_dir, histogram_freq=0, write_graph=True, write_images=False)
        self.model.fit_generator(
            train_data_gen,
            steps_per_epoch=train_data_gen.samples // self.batch_size,
            epochs=self.epochs,
            validation_data=valid_data_gen,
            validation_steps=valid_data_gen.samples // self.batch_size,
            callbacks=[cp_callback, tb_callback])

    def convert_checkpoint(self, final_checkpoint):
        checkpoint_dir = os.path.dirname(final_checkpoint)
        basename = os.path.basename(final_checkpoint).split(".")[0]
        save_path = os.path.join(checkpoint_dir, "tf_ckpt", "final_model.ckpt")
        # Add ops to save and restore all the variables.
        saver = tf.train.Saver()
        self.model.load_weights(final_checkpoint)
        sess = get_session()
        saver.save(sess, save_path)

    def save_frozen(self, frozen_filename):
        # First freeze the graph and remove training nodes.
        output_names = self.model.output.op.name
        sess = get_session()
        frozen_graph = tf.graph_util.convert_variables_to_constants(sess, sess.graph.as_graph_def(), [output_names])
        frozen_graph = tf.graph_util.remove_training_nodes(frozen_graph)
        # Save the model
        with open(frozen_filename, "wb") as ofile:
            ofile.write(frozen_graph.SerializeToString())

    def prediction(self, test_data_path):
        from PIL import Image
        import numpy as np
        test_images = glob(os.path.join(test_data_path, "*.jpg"))
        for impath in test_images:
            img = Image.open(impath)
            img = img.resize(self.input_shape[:2])
            img = np.expand_dims(np.array(img), axis=0) / 255.0
            output = self.model.predict(img, batch_size=1)
            print(output)
            pred = np.argmax(output)
            basename = os.path.basename(impath)
            if pred:
                print("{} : Prediction -  Dog".format(basename))
            else:
                print("{} : Prediction -  Cat".format(basename))


def test_case():
    try:
        model = Model()
        model.build_model()
        return True
    except Exception as e:
        print(e)
        return False


def test_case2():
    final_checkpoint = "/home/codesteller/workspace/ml_workspace/trt-custom-plugin/saved_model/" \
                       "checkpoints/saved_model-0001.h5"
    model = Model(input_shape=(150, 150, 3))
    model.build_model()
    model.convert_checkpoint(final_checkpoint)
    print("done")


def test_case3():
    final_checkpoint = "/home/codesteller/workspace/ml_workspace/trt_ws/trt-custom-plugin/saved_model/" \
                       "checkpoints/saved_model-0005.h5"
    test_data = "../test_data"
    cnn_model = Model(input_shape=(150, 150, 3))
    cnn_model.build_model()
    cnn_model.model.load_weights(final_checkpoint)
    # cnn_model.convert_checkpoint(final_checkpoint)
    cnn_model.prediction(test_data)


if __name__ == "__main__":
    # if test_case():
    #     print("Test Case Passed")
    # else:
    #     print("Test Case Passed")
    #
    # test_case2()
    test_case3()