以图搜图之模型篇： 基于 InceptionV3 的模型 finetune

# 基本流程# import osimport sysimport globimport argparseimport matplotlib.pyplot as pltfrom keras.applications.inception_v3 import InceptionV3, preprocess_inputfrom keras.models import Modelfrom keras.layers import Dense, GlobalAveragePooling2Dfrom keras.preprocessing.image import ImageDataGeneratorfrom keras.optimizers import SGD# 一、定义函数IM_WIDTH, IM_HEIGHT = 299, 299 ??# inceptionV3 指定图片尺寸FC_SIZE = 1024 ??????????????????# 全连接层的数量# 二、数据处理# 图片归类放在不同文件夹下train_dir = ‘E:/Project/Image/data/finetune/train‘ ?# 训练集数据val_dir = ‘E:/Project/Image/data/finetune/test‘ # 验证集数据nb_epoch = 1batch_size = 15nb_classes = len(glob.glob(train_dir + "/*")) ?# 分类数# 图片增强# ImageDataGenerator 会自动根据路径下的文件夹创建标签，所以在代码中只看到输入的 x, 看不到 ytrain_datagen = ImageDataGenerator( ???preprocessing_function=preprocess_input, ???rotation_range=30, ???width_shift_range=0.2, ???height_shift_range=0.2, ???shear_range=0.2, ???zoom_range=0.2, ???horizontal_flip=True)train_generator = train_datagen.flow_from_directory( ???train_dir, target_size=(IM_WIDTH, IM_HEIGHT),batch_size=batch_size, class_mode=‘categorical‘)validation_generator = train_datagen.flow_from_directory( ???val_dir, target_size=(IM_WIDTH, IM_HEIGHT),batch_size=batch_size, class_mode=‘categorical‘)# 三、使用 bottleneck finetune# 去掉 模型最外层的全连接层，添加上自己的 全连接层# 添加新层函数def add_new_last_layer(base_model, nb_classes): ???x = base_model.output ???x = GlobalAveragePooling2D()(x) # 下采样 ???x = Dense(FC_SIZE, activation=‘relu‘)(x) ???predict_bottle_feat = Dense(nb_classes, activation=‘softmax‘)(x) ???model = Model(input=base_model.input, output=predict_bottle_feat) ???return model# 冻结 base_model 所有层def setup_to_transfer_learn(model, base_model): ???for layer in base_model.layers: ???????layer.trainable = False ???model.compile(optimizer=‘rmsprop‘, loss=‘categorical_crossentropy‘, metrics=[‘accuracy‘])# 定义网络框架base_model = InceptionV3(weights=‘imagenet‘, include_top=False)model = add_new_last_layer(base_model, nb_classes)setup_to_transfer_learn(model, base_model)# 训练# 模式一训练steps = 20 # 可以自由定义，越大结果越精准，但过大容易过拟合history_tl = model.fit_generator( ?train_generator, ?epochs=nb_epoch, ?steps_per_epoch=steps, ?validation_data=validation_generator, ?validation_steps=steps, ?class_weight=‘auto‘)# 保存模型model.save("my_inceptionV3.h5")

from keras.preprocessing import imagefrom keras_applications.inception_v3 import preprocess_inputfrom keras.models import Model, load_modelimport numpy as nptarget_size = (229, 229) #fixed size for InceptionV3 architecturebase_model = load_model(filepath="my_inceptionV3.h5")# 需要提取那一层的特征，此处就写入指定层的名称model = Model(input=base_model.input, output=base_model.get_layer(‘block4_pool‘).output)img_path = "C:/Users/Administrator/Pictures/搜图/horse.jpg"img = image.load_img(img_path, target_size=target_size)x = image.img_to_array(img)x = np.expand_dims(x, axis=0)x = preprocess_input(x)block4_pool_features = model.predict(x)

from keras.preprocessing import imagefrom ?keras.models import load_modelimport numpy as npimport jsonfrom keras_applications.imagenet_utils import decode_predictionsdef predict(model, img, target_size): ?"""Run model prediction on image ?Args: ???model: keras model ???img: PIL format image ???target_size: (w,h) tuple ?Returns: ???list of predicted labels and their probabilities ?""" ?if img.size != target_size: ???img = img.resize(target_size) ?x = image.img_to_array(img) ?x = np.expand_dims(x, axis=0) ?x = preprocess_input(x) ?preds = model.predict(x) ??# 此处获取的为 ?return preds[0] ???????????# 返回 numpy array [classes, ]def decode_predict(probalities_list): ?with open("img_classes.json", ‘r‘) as load_f: ???load_dict = json.load(load_f) ?index = probalities_list.index(max(probalities_list)) ?target_class = load_dict[str(index)] ?return target_classtarget_size = (229, 229) #fixed size for InceptionV3 architecturemodel = load_model(filepath="my_inceptionV3.h5")img = image.load_img("C:/Users/Administrator/Pictures/搜图/horse.jpg")res_numpy = predict(model, img, target_size=target_size)res_list = res_numpy.tolist()target_class = decode_predict(res_list)print(target_class)