Created
October 22, 2017 15:59
-
-
Save ardamavi/f434c61aae7a0d599c3af089ce523d3e to your computer and use it in GitHub Desktop.
Deep Learning Models With Keras
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| # Arda Mavi | |
| import os | |
| import sys | |
| from keras.models import Model | |
| from keras.layers import Input, Conv2D, MaxPooling2D, Activation, Flatten, Dense, Dropout | |
| from keras.models import model_from_json | |
| def save_model(model, path='Data/Model'): | |
| if not os.path.exists(path): | |
| os.makedirs(path) | |
| model_json = model.to_json() | |
| with open(path+'/model.json', 'w') as model_file: | |
| model_file.write(model_json) | |
| # serialize weights to HDF5 | |
| model.save_weights(path+'/weights.h5') | |
| print('Model and weights saved') | |
| return | |
| def get_saved_model(model_path='Data/Model/model.json', weights_path='Data/Model/weights.h5'): | |
| # Getting model: | |
| model_file = open(model_path, 'r') | |
| model = model_file.read() | |
| model_file.close() | |
| model = model_from_json(model) | |
| # Getting weights | |
| model.load_weights(weights_path) | |
| return model | |
| def get_model(model_name='LENET-5'): | |
| if model_name == 'LENET-5': | |
| return get_lenet_5() | |
| elif model_name == 'MVGG-5': | |
| return get_mvgg_5() | |
| elif model_name == 'MVGG-6': | |
| return get_mvgg_6() | |
| elif model_name == 'MVGG-7': | |
| return get_mvgg_7() | |
| elif model_name == 'MVGG-8': | |
| return get_mvgg_8() | |
| elif model_name == 'MVGG-9': | |
| return get_mvgg_9() | |
| else: | |
| print('Model "{0}" not found!'.format(model_name)) | |
| return None | |
| def get_lenet_5(): | |
| inputs = Input(shape=(28, 28, 1)) | |
| conv_1 = Conv2D(6, (5, 5), strides=(1,1), padding='SAME')(inputs) | |
| act_1 = Activation('relu')(conv_1) | |
| pooling_1 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_1) | |
| conv_2 = Conv2D(16, (5, 5), strides=(1,1), padding='SAME')(pooling_1) | |
| act_2 = Activation('relu')(conv_2) | |
| pooling_2 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_2) | |
| conv_3 = Conv2D(120, (5, 5), strides=(1,1), padding='SAME')(pooling_2) | |
| act_3 = Activation('relu')(conv_3) | |
| flat_1 = Flatten()(act_3) | |
| fc = Dense(84)(flat_1) | |
| fc = Activation('relu')(fc) | |
| fc = Dropout(0.5)(fc) | |
| fc = Dense(10)(fc) | |
| outputs = Activation('softmax')(fc) | |
| model = Model(inputs=inputs, outputs=outputs) | |
| model.compile(loss='categorical_crossentropy', optimizer='adam') | |
| return model | |
| def get_mvgg_5(): | |
| inputs = Input(shape=(28, 28, 1)) | |
| conv_1 = Conv2D(16, (3, 3), strides=(1,1), padding='SAME')(inputs) | |
| act_1 = Activation('relu')(conv_1) | |
| conv_2 = Conv2D(16, (3, 3), strides=(1,1), padding='SAME')(act_1) | |
| act_2 = Activation('relu')(conv_2) | |
| pooling_1 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_2) | |
| conv_3 = Conv2D(48, (3, 3), strides=(1,1), padding='SAME')(pooling_1) | |
| act_3 = Activation('relu')(conv_3) | |
| pooling_2 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_3) | |
| flat_1 = Flatten()(pooling_2) | |
| fc = Dense(128)(flat_1) | |
| fc = Activation('relu')(fc) | |
| fc = Dropout(0.5)(fc) | |
| fc = Dense(10)(fc) | |
| outputs = Activation('softmax')(fc) | |
| model = Model(inputs=inputs, outputs=outputs) | |
| model.compile(loss='categorical_crossentropy', optimizer='adam') | |
| return model | |
| def get_mvgg_6(): | |
| inputs = Input(shape=(28, 28, 1)) | |
| conv_1 = Conv2D(16, (3, 3), strides=(1,1), padding='SAME')(inputs) | |
| act_1 = Activation('relu')(conv_1) | |
| conv_2 = Conv2D(16, (3, 3), strides=(1,1), padding='SAME')(act_1) | |
| act_2 = Activation('relu')(conv_2) | |
| pooling_1 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_2) | |
| conv_3 = Conv2D(32, (3, 3), strides=(1,1), padding='SAME')(pooling_1) | |
| act_3 = Activation('relu')(conv_3) | |
| conv_4 = Conv2D(48, (3, 3), strides=(1,1), padding='SAME')(act_3) | |
| act_4 = Activation('relu')(conv_4) | |
| pooling_2 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_4) | |
| flat_1 = Flatten()(pooling_2) | |
| fc = Dense(128)(flat_1) | |
| fc = Activation('relu')(fc) | |
| fc = Dropout(0.5)(fc) | |
| fc = Dense(10)(fc) | |
| outputs = Activation('softmax')(fc) | |
| model = Model(inputs=inputs, outputs=outputs) | |
| model.compile(loss='categorical_crossentropy', optimizer='adam') | |
| return model | |
| def get_mvgg_7(): | |
| inputs = Input(shape=(28, 28, 1)) | |
| conv_1 = Conv2D(16, (3, 3), strides=(1,1), padding='SAME')(inputs) | |
| act_1 = Activation('relu')(conv_1) | |
| conv_2 = Conv2D(16, (3, 3), strides=(1,1), padding='SAME')(act_1) | |
| act_2 = Activation('relu')(conv_2) | |
| pooling_1 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_2) | |
| conv_3 = Conv2D(32, (3, 3), strides=(1,1), padding='SAME')(pooling_1) | |
| act_3 = Activation('relu')(conv_3) | |
| conv_4 = Conv2D(32, (3, 3), strides=(1,1), padding='SAME')(act_3) | |
| act_4 = Activation('relu')(conv_4) | |
| pooling_2 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_4) | |
| conv_5 = Conv2D(48, (3, 3), strides=(1,1), padding='SAME')(pooling_2) | |
| act_5 = Activation('relu')(conv_5) | |
| pooling_3 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_5) | |
| flat_1 = Flatten()(pooling_3) | |
| fc = Dense(128)(flat_1) | |
| fc = Activation('relu')(fc) | |
| fc = Dropout(0.5)(fc) | |
| fc = Dense(10)(fc) | |
| outputs = Activation('softmax')(fc) | |
| model = Model(inputs=inputs, outputs=outputs) | |
| model.compile(loss='categorical_crossentropy', optimizer='adam') | |
| return model | |
| def get_mvgg_8(): | |
| inputs = Input(shape=(28, 28, 1)) | |
| conv_1 = Conv2D(16, (3, 3), strides=(1,1), padding='SAME')(inputs) | |
| act_1 = Activation('relu')(conv_1) | |
| conv_2 = Conv2D(16, (3, 3), strides=(1,1), padding='SAME')(act_1) | |
| act_2 = Activation('relu')(conv_2) | |
| pooling_1 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_2) | |
| conv_3 = Conv2D(32, (3, 3), strides=(1,1), padding='SAME')(pooling_1) | |
| act_3 = Activation('relu')(conv_3) | |
| conv_4 = Conv2D(32, (3, 3), strides=(1,1), padding='SAME')(act_3) | |
| act_4 = Activation('relu')(conv_4) | |
| pooling_2 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_4) | |
| conv_5 = Conv2D(48, (3, 3), strides=(1,1), padding='SAME')(pooling_2) | |
| act_5 = Activation('relu')(conv_5) | |
| conv_6 = Conv2D(48, (3, 3), strides=(1,1), padding='SAME')(act_5) | |
| act_6 = Activation('relu')(conv_6) | |
| pooling_3 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_6) | |
| flat_1 = Flatten()(pooling_3) | |
| fc = Dense(128)(flat_1) | |
| fc = Activation('relu')(fc) | |
| fc = Dropout(0.5)(fc) | |
| fc = Dense(10)(fc) | |
| outputs = Activation('softmax')(fc) | |
| model = Model(inputs=inputs, outputs=outputs) | |
| model.compile(loss='categorical_crossentropy', optimizer='adam') | |
| return model | |
| def get_mvgg_9(): | |
| inputs = Input(shape=(28, 28, 1)) | |
| conv_1 = Conv2D(16, (3, 3), strides=(1,1), padding='SAME')(inputs) | |
| act_1 = Activation('relu')(conv_1) | |
| conv_2 = Conv2D(16, (3, 3), strides=(1,1), padding='SAME')(act_1) | |
| act_2 = Activation('relu')(conv_2) | |
| pooling_1 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_2) | |
| conv_3 = Conv2D(32, (3, 3), strides=(1,1), padding='SAME')(pooling_1) | |
| act_3 = Activation('relu')(conv_3) | |
| conv_4 = Conv2D(32, (3, 3), strides=(1,1), padding='SAME')(act_3) | |
| act_4 = Activation('relu')(conv_4) | |
| pooling_2 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_4) | |
| conv_5 = Conv2D(48, (3, 3), strides=(1,1), padding='SAME')(pooling_2) | |
| act_5 = Activation('relu')(conv_5) | |
| conv_6 = Conv2D(48, (3, 3), strides=(1,1), padding='SAME')(act_5) | |
| act_6 = Activation('relu')(conv_6) | |
| pooling_3 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='SAME')(act_6) | |
| conv_7 = Conv2D(64, (3, 3), strides=(1,1), padding='SAME')(pooling_3) | |
| act_7 = Activation('relu')(conv_7) | |
| pooling_4 = MaxPooling2D(pool_size=(4, 4), strides=(4, 4), padding='SAME')(act_7) | |
| flat_1 = Flatten()(pooling_2) | |
| fc = Dense(128)(flat_1) | |
| fc = Activation('relu')(fc) | |
| fc = Dropout(0.5)(fc) | |
| fc = Dense(10)(fc) | |
| outputs = Activation('softmax')(fc) | |
| model = Model(inputs=inputs, outputs=outputs) | |
| model.compile(loss='categorical_crossentropy', optimizer='adam') | |
| return model | |
| if __name__ == '__main__': | |
| model_name = sys.argv[1] | |
| model = get_model(model_name) | |
| if model != None: | |
| print('Model Architecture:') | |
| print(model.summary()) | |
| save_model(model) | |
| print('Model saved!') |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Models:
