精度低 - 迁移学习 + 瓶颈 keras-tensorflow (resnet50)

Question

我正在尝试在 google Colaboratory 笔记本上使用 keras/tensorflow 进行迁移学习/瓶颈。我的问题是准确率不超过 6%（Kaggle 的狗品种挑战，120 类，数据由 datagen.flow_from_directory 生成）

下面是我的代码，有什么我遗漏的吗？

tr_model=ResNet50(include_top=False,
                  weights='imagenet',
                 input_shape = (224, 224, 3),)

datagen = ImageDataGenerator(rescale=1. / 255)

#### Training ####
train_generator = datagen.flow_from_directory(train_data_dir,
                                                    target_size=(image_size,image_size),
                                                    class_mode=None,
                                                    batch_size=batch_size,
                                                    shuffle=False)
bottleneck_features_train = tr_model.predict_generator(train_generator)
train_labels = to_categorical(train_generator.classes , num_classes=num_classes)

#### Validation ####
validation_generator = datagen.flow_from_directory(validation_data_dir, 
                                                    target_size=(image_size,image_size),
                                                    class_mode=None,
                                                    batch_size=batch_size,
                                                    shuffle=False)
bottleneck_features_validation = tr_model.predict_generator(validation_generator)
validation_labels = to_categorical(validation_generator.classes, num_classes=num_classes)

#### Model creation ####
model = Sequential()
model.add(Flatten(input_shape=bottleneck_features_train.shape[1:]))
model.add(Dense(num_class, activation='softmax'))

model.compile(optimizer='rmsprop',
              loss='categorical_crossentropy',
              metrics=['accuracy'])

history = model.fit(bottleneck_features_train, train_labels,
                    epochs=30,
                    batch_size=batch_size,
                    validation_data=(bottleneck_features_validation, validation_labels))

我得到一个 val_acc = 0.0592

当我在最后一层使用 ResNet50 时，我得到了 82% 的分数。

谁能发现我的代码有什么问题。

Answer 1

抑制重新缩放并添加预处理帮助很大。

这些修改有很大帮助：

from keras.applications.resnet50 import preprocess_input
datagen = ImageDataGenerator(preprocessing_function=preprocess_input)

我现在的准确率为 80%

精度低 - 迁移学习 + 瓶颈 keras-tensorflow (resnet50)

Low accurcacy - Transfer learning + bottle-neck keras-tensorflow (resnet50)

keras

tensorflow

google-colaboratory