使用 keras 输出图像 - 如何为输出层指定形状。 输入图像和输出图像尺寸为(210,210,3)

Question

I am new to Keras. I am trying to feed a color inverted image into a neural network and then predict the real image. So that my x becomes the inverted image and y becomes the real image. But I am not knowing how to get an output image with keras.

Here is my code.

import numpy as np
from keras.preprocessing.image import load_img
from keras.preprocessing.image import img_to_array
from keras.models import Sequential
from keras.layers import Conv2D,MaxPooling2D,Flatten,Dense

real=[]
for i in range(1,11):
    img=load_img(f"resized 1/{i}.jpg",target_size=(210,200))
    img=img_to_array(img)
    real.append(img)

invs=[]
for i in range(1,11):
    img=load_img(f"resized 2/{i}i.jpg",target_size=(210,200))
    img=img_to_array(img)
    invs.append(img)

x=np.array(invs)
y=np.array(real)
x=x/255
y=y/255


model=Sequential()

model.add(Conv2D(64,(3,3),activation='relu'))
model.add(MaxPooling2D((2,2)))

model.add(Conv2D(64,(3,3),activation='relu'))
model.add(MaxPooling2D((2,2)))

model.add(Flatten())
model.add(Dense(128,input_shape=x.shape[1:],activation='relu'))

what should i do after this how should i specify the output layer....I'm stuck

model.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])
model.fit(x,y,epochs=5,validation_split=0.1)

Thanks in advance.... I have referred this post but cant get anything - Keras: feed images into CNN and get image output

Answer 1

This is not a classification or regression problem. Thus, the type of neural network that you are using (Convolutional) is not suitable for this problem.

I think neural style transfer would be more appropriate. In this case Generative Adversarial Network (GAN) should be used.
You can read example about it here: https://www.tensorflow.org/tutorials/generative/dcgan

Answer 2

After doing research in GANs.This problem is suitable for Pix2Pix GAN.

https://machinelearningmastery.com/how-to-implement-pix2pix-gan-models-from-scratch-with-keras/