【发布时间】：2018-05-22 03:05:52
【问题描述】：
我的数据集如下： 训练集：5589 张图片 验证集：1398 张图片 测试集：1996 张图片 尺寸：1156,256,1 该问题是一个二元分类问题。我使用热编码的目标数组 [0,1],[1,0] 得到了一些结果（在测试集中达到了约 83% 的准确率）。意识到这是多么愚蠢，我将目标数组更改为二进制形式 [0] 或 1，并将 categorical_crossentropy 更改为二进制交叉熵。
使用这种方法，无论我使用什么学习率，验证准确率都会停留在 82.05%，而训练准确率会停留在 25.80%。当然，这没有任何意义，在测试集中的准确率约为 30%。
为什么会发生这种情况？我检查了训练数据和元数据，它们是正确的。我在下面发布我的代码。
inp = Input(shape=input_shape)
out = Conv2D(16, (5, 5),activation = 'relu', kernel_initializer='glorot_uniform', kernel_regularizer=regularizers.l2(0.01), padding='same')(inp)
out = MaxPooling2D(pool_size=(2, 2))(out)
out = Dropout(0.5)(out)
out = Conv2D(32, (3, 3),activation = 'relu',kernel_initializer='glorot_uniform',kernel_regularizer=regularizers.l2(0.01), padding='same')(out)
out = MaxPooling2D(pool_size=(2, 2))(out)
out = Dropout(0.5)(out)
out = Conv2D(32, (3, 3),activation = 'relu',kernel_initializer='glorot_uniform',kernel_regularizer=regularizers.l2(0.01), padding='same')(out)
out = Dropout(0.5)(out)
out = Conv2D(64, (3, 3), activation = 'relu',kernel_initializer='glorot_uniform',kernel_regularizer=regularizers.l2(0.01), padding='same')(out)
out = Conv2D(64, (3, 3),activation = 'relu', kernel_initializer='glorot_uniform',kernel_regularizer=regularizers.l2(0.01), padding='same')(out)
out = MaxPooling2D(pool_size=(2, 2))(out)
out = Conv2D(128, (3, 3), activation = 'relu',kernel_initializer='glorot_uniform',kernel_regularizer=regularizers.l2(0.01), padding='same')(out)
out = Conv2D(128, (3, 3),activation = 'relu', kernel_initializer='glorot_uniform',kernel_regularizer=regularizers.l2(0.01), padding='same')(out)
out = MaxPooling2D(pool_size=(2, 2))(out)
out = Conv2D(256, (3, 3),activation = 'relu', kernel_initializer='glorot_uniform',kernel_regularizer=regularizers.l2(0.01), padding='same')(out)
out = Conv2D(256, (3, 3), activation = 'relu',kernel_initializer='glorot_uniform',kernel_regularizer=regularizers.l2(0.01), padding='same')(out)
out = MaxPooling2D(pool_size=(2, 2))(out)
out = Conv2D(512, (3, 3), activation = 'relu',kernel_initializer='glorot_uniform', kernel_regularizer=regularizers.l2(0.01), padding='same')(out)
out = MaxPooling2D(pool_size=(2, 2))(out)
out = Flatten()(out)
out = Dropout(0.5)(out)
dense1 = Dense(1, activation="softmax")(out)
model = Model(inputs = inp, outputs = dense1)
时代看起来像这样： Epochs
【问题讨论】：
-
最后一个密集层
activation='sigmoid'
标签： machine-learning neural-network deep-learning keras conv-neural-network