代码
import numpy as np
from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential,Model
from keras.layers import LeakyReLU,Dropout, Flatten, Dense,Input
from keras import applications
from keras.preprocessing import image
from keras import backend as K
from keras import regularizers
from keras.optimizers import adam
K.set_image_dim_ordering('tf')
input_tensor = Input(shape=(150,150,3))
img_width, img_height = 150,150
top_model_weights_path = 'bottleneck_fc_model.h5'
train_data_dir = 'Cats and Dogs Dataset/train'
validation_data_dir = 'Cats and Dogs Dataset/validation'
nb_train_samples = 20000
nb_validation_samples = 5000
epochs = 50
batch_size = 128
base_model=applications.inception_v3.InceptionV3(include_top=False, weights='imagenet', input_tensor=input_tensor, pooling=None)
i=0;
for layer in base_model.layers:
layer.trainable = False
i+=1
base_model.output
top_model=Sequential()
top_model.add(Flatten(input_shape=base_model.output_shape[1:]))
top_model.add(Dense(1024,activation="relu"))
top_model.add(Dropout(0.5))
top_model.add(Dense(10,activation="relu"))//Layer with issue
top_model.add(Dropout(0.8))//
top_model.add(Dense(2, activation='softmax'))
model = Model(inputs=base_model.input,outputs=top_model(base_model.output))
model.summary
datagen = ImageDataGenerator(rescale=1. / 255)
train_data = datagen.flow_from_directory(train_data_dir,target_size=(img_width, img_height),batch_size=batch_size,classes=[ 'cats','dogs'])#,class_mode="binary",shuffle=True)
validation_data = datagen.flow_from_directory(validation_data_dir,target_size=(img_width, img_height), batch_size=batch_size,classes=['cats','dogs'])#,class_mode="binary",shuffle=True)
adm=adam(lr=0.02)
model.compile(optimizer=adm,loss='categorical_crossentropy', metrics=['accuracy'])
model.fit_generator(train_data, steps_per_epoch=nb_train_samples//batch_size, epochs=epochs,validation_data=validation_data, shuffle=True,verbose=1)
我已经使用 keras（使用初始网络学习传输）在猫狗数据集（https://www.kaggle.com/c/dogs-vs-cats/data ）上实现了图像分类器。代码运行没有错误，但从第一个 epoch 开始，验证集和训练集的准确率停留在 50%，并且损失没有减少。我正在使用带有氢的原子。
当我删除标记层时，问题就消失了，我似乎无法理解为什么会这样。我试图解决这个问题
- 不同的批次大小 - 4,16,64,256
- 更改优化器 - 尝试使用修改后的学习率的 adam、rmsprop、sgd
- 尝试了不同的激活层——relu、sigmoid 和leakyrelu
- 更改了 dropout - 当 dropout 为 0.9 时问题消失（即使图层无用，这显然是有效的，但也指出我缺少一些东西）
- 将最终激活更改为 sigmoid
有人可以告诉我我缺少什么，因为我想不出添加图层停止学习的任何原因