重量没有改变,而在初期V3训练,但在VGG16工作良好

def Inception():

    model = inception_v3.InceptionV3(include_top=False, weights='imagenet',
                  input_shape=(imageheight, imagewidth, 3))
    x = model.output
    x = Flatten()(x)
    x = Dense(1, activation='softmax', name='predictions')(x)
    model = Model(input=model.input, output=x)

    # print(model.summary())

    return model

我尝试过用不同的值初始化权重,甚至尝试过不同的优化器:

sgd = optimizers.SGD(lr = 5e-5, decay=1e-6, momentum=0.9, nesterov=True) 

我得到的结果是:

Epoch 1/100
611/611 [==============================] - 14s 23ms/step - loss: 11.5589 - acc: 0.2750 - val_loss: 11.5660 - val_acc: 0.2745
Epoch 2/100
611/611 [==============================] - 6s 9ms/step - loss: 11.5589 - acc: 0.2750 - val_loss: 11.5660 - val_acc: 0.2745
Epoch 3/100
611/611 [==============================] - 6s 9ms/step - loss: 11.5589 - acc: 0.2750 - val_loss: 11.5660 - val_acc: 0.2745
Epoch 4/100
611/611 [==============================] - 6s 9ms/step - loss: 11.5589 - acc: 0.2750 - val_loss: 11.5660 - val_acc: 0.2745

奇怪的是,当我在VGG16上使用相同的数据集时,它就工作了。唯一的问题是我从头开始为VGG16编写了体系结构。