Keras CNN 精度高但预测不佳。 如何改进它

Question

I have to identify a retinal disease with CNN. I have 1400 images, 700 each class. My classes are (0 - no PDR) and (1 - PDR). I'm trying to make a model to identify if an input retina have the disease in level 4 or not.

I'm making a follow manipulation with my images and rezise all to 256x256:

ImageCV[index] = cv2.addWeighted(ImageCV[index],4, cv2.GaussianBlur(ImageCV[index],(0,0), 256/30), -4, 128)

And it made the follow with my imgs: https://imgur.com/X1p9G1c

Then, when I train my model I got a very high accuracy (like 99....) but when I try to predict some test images, it fails.. for example, I've putted 10 PDR examples in test folder and tries to predict them (all must be 1).. this is the result:

[[0.]]
[[0.]]
[[1.]]
[[0.]]
[[0.]]
[[0.]]
[[1.]]
[[0.]]
[[0.]]
[[0.]]

This is my model:

visible = Input(shape=(256,256,3))
conv1 = Conv2D(16, kernel_size=(3,3), activation='relu', strides=(1, 1))(visible)
conv2 = Conv2D(16, kernel_size=(3,3), activation='relu', strides=(1, 1))(conv1)
bat1 = BatchNormalization()(conv2)
conv3 = ZeroPadding2D(padding=(1, 1))(bat1)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv3)

conv4 = Conv2D(32, kernel_size=(3,3), activation='relu', padding='valid', kernel_regularizer=regularizers.l2(0.01))(pool1)
conv5 = Conv2D(32, kernel_size=(3,3), activation='relu', padding='valid', kernel_regularizer=regularizers.l2(0.01))(conv4)
bat2 = BatchNormalization()(conv5)
pool2 = MaxPooling2D(pool_size=(1, 1))(bat2)

conv6 = Conv2D(64, kernel_size=(3,3), activation='relu',strides=(1, 1), padding='valid')(pool2)
conv7 = Conv2D(64, kernel_size=(3,3), activation='relu',strides=(1, 1), padding='valid')(conv6)
bat3 = BatchNormalization()(conv7)
conv7 = ZeroPadding2D(padding=(1, 1))(bat3)
pool3 = MaxPooling2D(pool_size=(1, 1))(conv7)

conv8 = Conv2D(128, kernel_size=(3,3), activation='relu', padding='valid', kernel_regularizer=regularizers.l2(0.01))(pool3)
conv9 = Conv2D(128, kernel_size=(2,2), activation='relu', strides=(1, 1), padding='valid')(conv8)
bat4 = BatchNormalization()(conv9)
pool4 = MaxPooling2D(pool_size=(1, 1))(bat4)

flat = Flatten()(pool4)

output = Dense(1, activation='sigmoid')(flat)
model = Model(inputs=visible, outputs=output)

opt = optimizers.adam(lr=0.001, decay=0.0)

model.compile(optimizer= opt, loss='binary_crossentropy', metrics=['accuracy'])


data, labels = ReadImages(TRAIN_DIR)

test, lt = ReadImages(TEST_DIR)

data = np.array(data)
labels = np.array(labels)
test = np.array(test)
lt = np.array(lt)

np.random.permutation(len(data))
np.random.permutation(len(labels))
np.random.permutation(len(test))
np.random.permutation(len(lt))

model.fit(data, labels, epochs=7, validation_data = (test,lt))

model.save('model.h5')

And this is predict.py

model = load_model('model.h5')

for filename in os.listdir(r'v/'):
    if filename.endswith(".jpg") or filename.endswith(".ppm") or filename.endswith(".jpeg"):
        ImageCV = cv2.resize(cv2.imread(os.path.join(TEST_DIR) + filename), (256,256))
        ImageCV = cv2.addWeighted(ImageCV,4, cv2.GaussianBlur(ImageCV,(0,0), 256/30), -4, 128)
        cv2.imshow('image', ImageCV)
        cv2.waitKey(0)
        cv2.destroyAllWindows()
        ImageCV = ImageCV.reshape(-1,256,256,3)
        print(model.predict(ImageCV))

What do I could to do to improve my predictions at all? I strongly appreciate your help

UPDATE Well, I tried to do all of were said in answers but still'nt working... this is my code now:

visible = Input(shape=(256,256,3))
conv1 = Conv2D(16, kernel_size=(3,3), activation='relu', strides=(1, 1))(visible)
conv2 = Conv2D(32, kernel_size=(3,3), activation='relu', strides=(1, 1))(conv1)
bat1 = BatchNormalization()(conv2)
conv3 = ZeroPadding2D(padding=(1, 1))(bat1)
pool1 = MaxPooling2D(pool_size=(2, 2))(conv3)
drop1 = Dropout(0.30)(pool1)

conv4 = Conv2D(32, kernel_size=(3,3), activation='relu', padding='valid', kernel_regularizer=regularizers.l2(0.01))(drop1)
conv5 = Conv2D(64, kernel_size=(3,3), activation='relu', padding='valid', kernel_regularizer=regularizers.l2(0.01))(conv4)
bat2 = BatchNormalization()(conv5)
pool2 = MaxPooling2D(pool_size=(1, 1))(bat2)
drop1 = Dropout(0.30)(pool2)

conv6 = Conv2D(128, kernel_size=(3,3), activation='relu', padding='valid', kernel_regularizer=regularizers.l2(0.01))(pool2)
conv7 = Conv2D(128, kernel_size=(2,2), activation='relu', strides=(1, 1), padding='valid')(conv6)
bat3 = BatchNormalization()(conv7)
pool3 = MaxPooling2D(pool_size=(1, 1))(bat3)
drop1 = Dropout(0.30)(pool3)

flat = Flatten()(pool3)
drop4 = Dropout(0.50)(flat)

output = Dense(1, activation='sigmoid')(drop4)
model = Model(inputs=visible, outputs=output)

opt = optimizers.adam(lr=0.001, decay=0.0)

model.compile(optimizer= opt, loss='binary_crossentropy', metrics=['accuracy'])

data, labels = ReadImages(TRAIN_DIR)
test, lt = ReadImages(TEST_DIR)

data = np.array(data)
labels = np.array(labels)

perm = np.random.permutation(len(data))
data = data[perm]
labels = labels[perm]
#model.fit(data, labels, epochs=8, validation_data = (np.array(test), np.array(lt)))

aug = ImageDataGenerator(rotation_range=20, width_shift_range=0.2, height_shift_range=0.2, shear_range=0.15,
    horizontal_flip=True)

# train the network
model.fit_generator(aug.flow(data, labels, batch_size=32),
    validation_data=(np.array(test), np.array(lt)), steps_per_epoch=len(data) // 32,
    epochs=7)

And this is the return:

Epoch 1/7
43/43 [==============================] - 1004s 23s/step - loss: 1.8090 - acc: 0.9724 - val_loss: 1.7871 - val_acc: 0.9861
Epoch 2/7
43/43 [==============================] - 1003s 23s/step - loss: 1.8449 - acc: 0.9801 - val_loss: 1.4828 - val_acc: 1.0000
Epoch 3/7
43/43 [==============================] - 1092s 25s/step - loss: 1.5704 - acc: 0.9920 - val_loss: 1.3985 - val_acc: 1.0000
Epoch 4/7
43/43 [==============================] - 1062s 25s/step - loss: 1.5219 - acc: 0.9898 - val_loss: 1.3167 - val_acc: 1.0000
Epoch 5/7
43/43 [==============================] - 990s 23s/step - loss: 2.5744 - acc: 0.9222 - val_loss: 2.9347 - val_acc: 0.9028
Epoch 6/7
43/43 [==============================] - 983s 23s/step - loss: 1.6053 - acc: 0.9840 - val_loss: 1.3299 - val_acc: 1.0000
Epoch 7/7
43/43 [==============================] - 974s 23s/step - loss: 1.6180 - acc: 0.9801 - val_loss: 1.5181 - val_acc: 0.9861

I'd have added dropouts, reduce the model layers, put data augmentation, and doesn't work at all (all the predictions returns 0)...

Please anyone can help in it.

Answer 1

It seems you having issues with overfitting. I have kind of a personal dilemma in here whether this is off-topic or not because the approach one can give it's somewhat opinion-based, but here I go: First off, if you need to regularize an overfitting network, you want to use dropout starting at 0.25 and checking if that improves the model. Data augmentation is a must when dealing with overfitting, together with batch normalization (which you are applying). If this still not solves your overfitting problems, then you should try working on your network architecture to tweak it in order to make it generalize better. Did you check a simple of the inputs being used for train and test?

TLDR: Try dropout and data augmentation, if it doesn't work and your data is correct, you might have to work on improving the architecture to make a better generalizing model.

EDIT: The consensus on approaching this kind of models is to have it overfit at first, with a decent accuracy and then work towards generalizing it without losing accuracy if possible.

Answer 2

Your model is very big for your training set size, leading to overfitting. Use less layers

Answer 3

You're overfitting ie have a high variance problem. Try adding some soft dropout (0.2-0.3) at the end of your convolution blocks. You can also add a couple of Dense layers with decreasing unit numbers before the output layer and put more dropout layers in between those (0.5+).

You should also implement more data augmentation such as rotations, flips, random noise, random brightness, etc. Check out Keras' documentation for the ImageDataGenerator class.