keras预测内存交换无限增加

Question

我使用keras实现了一个分类程序.我有一大堆图像,我想使用for循环预测每个图像.

但是,每次计算新图像时,交换内存都会增加.我试图删除预测函数中的所有变量(我确信它在此函数内部存在问题)但内存仍然增加.

for img in images:
    predict(img, model, categ_par, gl_par)

和相应的功能:

def predict(image_path, model, categ_par, gl_par):   
    print("[INFO] loading and preprocessing image...")

    orig = cv2.imread(image_path)  

    image = load_img(image_path, target_size=(gl_par.img_width, gl_par.img_height))  
    image = img_to_array(image)  

    # important! otherwise the predictions will be '0'  
    image = image / 255  

    image = np.expand_dims(image, axis=0)

    # build the VGG16 network
    if(categ_par.method == 'VGG16'):
        model = applications.VGG16(include_top=False, weights='imagenet')  

    if(categ_par.method == 'InceptionV3'):
        model = applications.InceptionV3(include_top=False, weights='imagenet')  

    # get the bottleneck prediction from the pre-trained VGG16 model  
    bottleneck_prediction = model.predict(image)  

    # build top model  
    model = Sequential()  
    model.add(Flatten(input_shape=bottleneck_prediction.shape[1:]))  
    model.add(Dense(256, activation='relu'))  
    model.add(Dropout(0.5))  
    model.add(Dense(categ_par.n_class, activation='softmax'))  

    model.load_weights(categ_par.top_model_weights_path)  

    # use the bottleneck prediction on the top model to get the final classification  
    class_predicted = model.predict_classes(bottleneck_prediction) 
    probability_predicted = (model.predict_proba(bottleneck_prediction))

    classe = pd.DataFrame(list(zip(categ_par.class_indices.keys(), list(probability_predicted[0])))).\
    rename(columns = {0:'type', 1: 'prob'}).reset_index(drop=True)
    #print(classe)
    del model
    del bottleneck_prediction
    del image
    del orig
    del class_predicted
    del probability_predicted

    return classe.set_index(['type']).T

Answer 1

如果您正在使用TensorFlow后端,那么您将在for循环中为每个img构建一个模型.TensorFlow只是将图形附加到图形等上,这意味着内存刚刚上升.这是一个众所周知的事件,必须在超参数优化期间处理,当你要构建许多模型时,还要在这里.

from keras import backend as K

并把它放在predict()的末尾:

K.clear_session()

或者您可以构建一个模型并将其作为预测函数的输入提供,这样您每次都不会构建新模型.