在保存的 keras Inception v3 模型上开始迁移学习

原文标题 ：Start transfer learning on a saved keras Inception v3 model

我有一个经过训练的 Keras Inception v3 模型，名为“nsfw.299×299.h5”，我从 Github.Github 链接下载：https://github.com/GantMan/nsfw_model 该模型将图像分为以下 5 类：

1. 中性的
2. A片
3. 无尽的
4. 性感的
5. 绘画

该模型在一些咖啡杯图像上给出了误报并将它们分类为“色情”当它应该被归类为“中性”时。所以为了消除偏见，我下载了大约 400 张咖啡杯图像并下载了相同数量的图像对于其他课程也是如此，并且想再次训练这个模型。我该如何进行训练？

下面是模型摘要的最后几行。

 __________________________________________________________________________________________________
Model: "model"
__________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to                     
==================================================================================================        
                    

__________________________________________________________________________________________________
activation_94 (Activation)      (None, 8, 8, 192)    0           batch_normalization_94[0][0]     
__________________________________________________________________________________________________
mixed10 (Concatenate)           (None, 8, 8, 2048)   0           activation_86[0][0]              
                                                                 mixed9_1[0][0]                   
                                                                 concatenate_2[0][0]              
                                                                 activation_94[0][0]              
__________________________________________________________________________________________________
average_pooling2d_10 (AveragePo (None, 1, 1, 2048)   0           mixed10[0][0]                    
__________________________________________________________________________________________________
flatten_1 (Flatten)             (None, 2048)         0           average_pooling2d_10[0][0]       
__________________________________________________________________________________________________
dense_1 (Dense)                 (None, 256)          524544      flatten_1[0][0]                  
__________________________________________________________________________________________________
dropout_2 (Dropout)             (None, 256)          0           dense_1[0][0]                    
__________________________________________________________________________________________________
dense_2 (Dense)                 (None, 128)          32896       dropout_2[0][0]                  
__________________________________________________________________________________________________
dropout_3 (Dropout)             (None, 128)          0           dense_2[0][0]                    
__________________________________________________________________________________________________
dense (Dense)                   (None, 5)            645         dropout_3[0][0]                  
==================================================================================================
Total params: 22,360,869
Trainable params: 17,076,261
Non-trainable params: 5,284,608

下面给出了用于训练该模型的代码，我从同一个 Github 存储库中获取了该代码，我从以下位置下载了模型：https://github.com/GantMan/nsfw_model/blob/master/tf1/training/inceptionv3_transfer /train_initialization.py

conv_base = InceptionV3(
    weights='imagenet', 
    include_top=False, 
    input_shape=(height, width, constants.NUM_CHANNELS)
)

# First time run, no unlocking
conv_base.trainable = False

# Let's see it
print('Summary')
print(conv_base.summary())

# Let's construct that top layer replacement
x = conv_base.output
x = AveragePooling2D(pool_size=(8, 8))(x)
x - Dropout(0.4)(x)
x = Flatten()(x)
x = Dense(256, activation='relu', kernel_initializer=initializers.he_normal(seed=None), kernel_regularizer=regularizers.l2(.0005))(x)
x = Dropout(0.5)(x)
# Essential to have another layer for better accuracy
x = Dense(128,activation='relu', kernel_initializer=initializers.he_normal(seed=None))(x)
x = Dropout(0.25)(x)
predictions = Dense(constants.NUM_CLASSES,  kernel_initializer="glorot_uniform", activation='softmax')(x)

print('Stacking New Layers')
model = Model(inputs = conv_base.input, outputs=predictions)

# Load checkpoint if one is found
if os.path.exists(weights_file):
        print ("loading ", weights_file)
        model.load_weights(weights_file)

# Get all model callbacks
callbacks_list = callbacks.make_callbacks(weights_file)

print('Compile model')
# originally adam, but research says SGD with scheduler
# opt = Adam(lr=0.001, amsgrad=True)
opt = SGD(momentum=.9)
model.compile(
    loss='categorical_crossentropy',
    optimizer=opt,
    metrics=['accuracy']
)

# Get training/validation data via generators
train_generator, validation_generator = generators.create_generators(height, width)

print('Start training!')
history = model.fit_generator(
    train_generator,
    callbacks=callbacks_list,
    epochs=constants.TOTAL_EPOCHS,
    steps_per_epoch=constants.STEPS_PER_EPOCH,
    shuffle=True,
    workers=4,
    use_multiprocessing=False,
    validation_data=validation_generator,
    validation_steps=constants.VALIDATION_STEPS
)

# Save it for later
print('Saving Model')
model.save("nsfw." + str(width) + "x" + str(height) + ".h5")