问题
I am new to python and Keras. I am trying to do a binary classification using transfer learning from ResNet. My dataset is very small but I am using image augmentation. My cross-validation accuracy is just either of 2 values 0.3442 and 0.6558 for all images. Can anyone tell me why this happens? Also when I predict (0 or 1), it labels all images as one class(0). Here is my code:
from keras.preprocessing.image import ImageDataGenerator, load_img
from keras.models import Sequential,Model,load_model
from keras.layers import Conv2D, MaxPooling2D
from keras.layers import Activation, Dropout, Flatten, Dense, GlobalMaxPooling2D
from keras import backend as K
from keras.callbacks import TensorBoard
from keras.applications.resnet50 import ResNet50
from keras.optimizers import SGD, Adam
from keras.utils import plot_model
import matplotlib.pyplot as plt
import os, os.path
import glob
import cv2
import time
from keras.utils import np_utils
from keras.callbacks import ReduceLROnPlateau, CSVLogger, EarlyStopping
import numpy as np
import pydot
import graphviz
batch_size = 32
nb_classes = 2
data_augmentation = True
img_rows, img_cols = 224,224
img_channels = 3
#Creating array of training samples
train_path = "D:/data/train\*.*"
training_data=[]
for file in glob.glob(train_path):
print(file)
train_array= cv2.imread(file)
train_array=cv2.resize(train_array,(img_rows,img_cols),3)
training_data.append(train_array)
x_train=np.array(training_data)
#Creating array of validation samples
valid_path = "D:/data/valid\*.*"
valid_data=[]
for file in glob.glob(valid_path):
print(file)
valid_array= cv2.imread(file)
valid_array=cv2.resize(valid_array,(img_rows,img_cols),3)
valid_data.append(train_array)
x_valid=np.array(valid_data)
x_train = np.array(x_train, dtype="float")/255.0
x_valid = np.array(x_valid, dtype="float")/255.0
#Creating array for Labels
y_train=np.ones((num_trainsamples,),dtype = int)
y_train[0:224]=0 #Class1=0
y_train[225:363]=1 #Class2=1
print(y_train)
y_valid=np.ones((num_validsamples,),dtype = int)
y_valid[0:101]=0
y_valid[102:155]=1
print(y_valid)
y_train = np_utils.to_categorical(y_train,nb_classes,dtype='int32')
y_valid = np_utils.to_categorical(y_valid,nb_classes,dtype='int32')
base_model=ResNet50(weights='imagenet',include_top=False)
x = base_model.output
x = GlobalMaxPooling2D()(x)
x=Dense(1024,activation='relu')(x)
x=Dense(1024,activation='relu')(x)
x=Dense(512,activation='relu')(x)
x=Dense(2, activation= 'sigmoid')(x)
model = Model(inputs = base_model.input, outputs = x)
for i,layer in enumerate(model.layers):
print(i,layer.name)
for layer in model.layers[:75]:
layer.trainable=False
for layer in model.layers[75:]:
layer.trainable=True
adam = Adam(lr=0.0001)
model.compile(optimizer= adam, loss='binary_crossentropy', metrics=['accuracy'])
train_datagen = ImageDataGenerator(
brightness_range=(0.2,2.5),
rotation_range=180,
zoom_range=0.5,
width_shift_range=0.2,
height_shift_range=0.2,
horizontal_flip=True,
vertical_flip=True)
train_datagen.fit(x_train)
history= model.fit_generator(train_datagen.flow(x_train, y_train, batch_size = 10,shuffle=True),steps_per_epoch=len(x_train),epochs = 500,shuffle=True,
validation_data=(x_valid,y_valid),validation_steps=num_validsamples // batch_size,callbacks=[tensorboard])
eval = model.evaluate(x_valid, y_valid)
print ("Loss = " + str(eval[0]))
print ("Test Accuracy = " + str(eval[1]))
predictions= model.predict(x_valid)
print(predictions)
The training result is as follows:
Epoch 1/500
362/362 [==============================] - 34s 93ms/step - loss: 0.6060 - acc: 0.7257 - val_loss: 0.7747 - val_acc: 0.3442
Epoch 2/500
362/362 [==============================] - 30s 82ms/step - loss: 0.4353 - acc: 0.7722 - val_loss: 0.7658 - val_acc: 0.5000
Epoch 3/500
362/362 [==============================] - 30s 82ms/step - loss: 0.4391 - acc: 0.7863 - val_loss: 0.7949 - val_acc: 0.3442
Epoch 4/500
362/362 [==============================] - 30s 82ms/step - loss: 0.4007 - acc: 0.7992 - val_loss: 0.6540 - val_acc: 0.6558
Epoch 5/500
362/362 [==============================] - 30s 82ms/step - loss: 0.3638 - acc: 0.8226 - val_loss: 0.6460 - val_acc: 0.6558
Epoch 6/500
362/362 [==============================] - 30s 82ms/step - loss: 0.3509 - acc: 0.8294 - val_loss: 0.7875 - val_acc: 0.3442
Epoch 7/500
362/362 [==============================] - 30s 82ms/step - loss: 0.3406 - acc: 0.8359 - val_loss: 0.7667 - val_acc: 0.3442
Epoch 8/500
362/362 [==============================] - 29s 80ms/step - loss: 0.3410 - acc: 0.8365 - val_loss: 0.6900 - val_acc: 0.6558
Epoch 9/500
362/362 [==============================] - 29s 80ms/step - loss: 0.3297 - acc: 0.8366 - val_loss: 0.7292 - val_acc: 0.3442
Epoch 10/500
362/362 [==============================] - 29s 80ms/step - loss: 0.3262 - acc: 0.8412 - val_loss: 0.6829 - val_acc: 0.6558
Epoch 11/500
362/362 [==============================] - 29s 80ms/step - loss: 0.3168 - acc: 0.8457 - val_loss: 0.7032 - val_acc: 0.3442
Epoch 12/500
362/362 [==============================] - 29s 80ms/step - loss: 0.3195 - acc: 0.8452 - val_loss: 0.6985 - val_acc: 0.5000
Epoch 13/500
362/362 [==============================] - 29s 80ms/step - loss: 0.3030 - acc: 0.8432 - val_loss: 0.6740 - val_acc: 0.6558
Epoch 14/500
362/362 [==============================] - 29s 80ms/step - loss: 0.3191 - acc: 0.8405 - val_loss: 0.6896 - val_acc: 0.6558
Epoch 15/500
362/362 [==============================] - 29s 80ms/step - loss: 0.3084 - acc: 0.8437 - val_loss: 0.7114 - val_acc: 0.3442
回答1:
When you implement binary classification with any CNN model, you had better use a single unit in the last dense layer.
x=Dense(2, activation= 'sigmoid')(x) ---> x=Dense(1, activation= 'sigmoid')(x)
It will give you better performance than now.
来源：https://stackoverflow.com/questions/54797065/resnet-for-binary-classification-just-2-values-of-cross-validation-accuracy