undergraduate_project_ipynb.py

# -*- coding: utf-8 -*-
"""「UnderGraduate_Project.ipynb」的副本

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1hQ7f8_-LfqFYSIW6CuCY1Pttz0Oe5ESg
"""

!nvidia-smi

from google.colab import drive
drive.mount('/content/drive')

# Commented out IPython magic to ensure Python compatibility.
import numpy as np
from tensorflow.keras import layers
from keras.preprocessing import image
import tensorflow as tf 
from keras.models import Model,load_model
import os
import keras
from keras.callbacks import ModelCheckpoint,ReduceLROnPlateau
from keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPool2D, BatchNormalization
from keras.callbacks import ModelCheckpoint,ReduceLROnPlateau, EarlyStopping
from keras.layers import Lambda
from keras.optimizers import Adam, SGD
from keras.preprocessing.image import ImageDataGenerator
from keras import backend as K
import random
from PIL import Image 
from random import shuffle
# %load_ext tensorboard
import datetime

"""**Prepare**"""

import numpy as np
from numpy import genfromtxt
from numpy import asarray
import math
import copy
import os
from PIL import Image 
import cv2
import shutil

patch_size = 64 #input = 64x64
label_size = 128 #output = 128x128

#get RGGB bayer image
def bayer_reverse(img):
    height,width,c = img.shape;
    tmp = np.zeros([height,width]);
    for i in range( height ):
        for j in range( width ):
            if i % 2 == 0 :
                if j % 2 == 0:
                    tmp[i][j] = img[i][j][0];#R
                else:
                    tmp[i][j] = img[i][j][1];#G
            else :
                if j % 2 == 0:
                    tmp[i][j] = img[i][j][1];#G
                else:
                    tmp[i][j] = img[i][j][2];#B

    return tmp;

#split image to prepare the train set
def split(img,name,dir_path):
    height,width,c = img.shape;    
    count = 0;

    for i in range(0, height, 30):
        for j in range(0, width, 30):
            if( i + label_size < height and j + label_size < width ):                
                label = np.zeros([label_size,label_size,3])                
                tmp2  = np.zeros([label_size,label_size,3])                
                tmp2 = img[ i : i + label_size, j : j + label_size,:]
                label_img = Image.fromarray(tmp2)

                label[:,:,0] = tmp2[:,:,2]
                label[:,:,1] = tmp2[:,:,1]
                label[:,:,2] = tmp2[:,:,0]
                label_path = os.path.join(dir_path,'label/'+name.split('.')[0] +'_'+str(count)+'.png')                                                
                cv2.imwrite(label_path,label)

                
                zoom = label_img.resize((patch_size,patch_size), Image.BICUBIC)                 
                tmp3 = np.zeros([patch_size,patch_size])
                patch = np.zeros([patch_size,patch_size])

                zoom2 = np.array(zoom)                
                patch = bayer_reverse(zoom2)                
                                           
                patch_path = os.path.join(dir_path,'patch/'+name.split('.')[0] +'_'+str(count)+'.png')                          
                cv2.imwrite(patch_path, patch)

                count = count + 1    

def main():
    path = 'drive/MyDrive/Colab Notebooks/undergraduate_project'

    if os.path.exists(os.path.join(path,'patch')):
      shutil.rmtree(os.path.join(path,'patch'))
    os.makedirs(os.path.join(path,'patch'))

    if os.path.exists(os.path.join(path,'label')):
      shutil.rmtree(os.path.join(path,'label'))
    os.makedirs(os.path.join(path,'label'))


    dataset_path = os.path.join(path,'BSD200')
    entries = os.listdir(dataset_path)
    for entry in entries:
        print(entry)  
        img_path = dataset_path + '/' + entry
        img = Image.open(img_path)
        img = np.array(img)   
        split(img,entry,path)



if __name__ == '__main__':
    main()

"""**Training**"""

import numpy as np
from tensorflow.keras import layers
from keras.preprocessing import image
import tensorflow as tf 
from keras.models import Model,load_model
import os
import keras
from keras.callbacks import ModelCheckpoint,ReduceLROnPlateau, EarlyStopping
from keras.layers import Lambda
from keras.optimizers import Adam
from keras import backend as K
import random
from PIL import Image 
from random import shuffle

train_image = []
train_label = []

dir_path = 'drive/MyDrive/Colab Notebooks/undergraduate_project'
patch_path = os.path.join(dir_path,'patch')
entries = os.listdir(patch_path)
for entry in entries:  
  im = Image.open(patch_path+'/'+entry)
  img = image.img_to_array(im)
  # print(img.shape)
  # img = img/255.
  train_image.append(img)
train_image= np.stack(train_image)

print(train_image.shape)
  

label_path = os.path.join(dir_path,'label')
entries = os.listdir(label_path)
for entry in entries:  
  im = Image.open(label_path+'/'+entry)
  img = image.img_to_array(im)
  # img = img/255.
  train_label.append(img)
train_label = np.stack(train_label)

print(train_label.shape)



index = [i for i in range(train_image.shape[0])]
shuffle(index)
train_image = train_image[index,:,:,:];
train_label = train_label[index,:,:,:];

# np.save('drive/My Drive/Colab Notebooks/undergraduate_project/train_image.npy', train_image)
# np.save('drive/My Drive/Colab Notebooks/undergraduate_project/train_label.npy', train_label)

############################# Model Structure ################################################
def create_model():
  inputs = keras.Input(shape=(None,None,1))

  ini = keras.layers.Conv2D(filters = 128, #feature map number
                     kernel_size = 3, 
                     strides = 1,  # 2                     
                     padding = 'same', 
                     input_shape = (None,None,1))(inputs)

  x = keras.layers.PReLU(alpha_initializer='zeros', alpha_regularizer=None, alpha_constraint=None, shared_axes=[1,2])(ini)
  
  ini = keras.layers.Conv2D(filters = 128, #feature map number
                     kernel_size = 3, 
                     strides = 1,  # 2
                     activation = 'relu',
                     padding = 'same', 
                     input_shape = (None,None,128))(x)

  x = keras.layers.PReLU(alpha_initializer='zeros', alpha_regularizer=None, alpha_constraint=None, shared_axes=[1,2])(ini)

  ini = keras.layers.Conv2D(filters = 64, #feature map number
                     kernel_size = 3, 
                     strides = 1,  # 2
                     activation = 'relu',
                     padding = 'same', 
                     input_shape = (None,None,128))(x)

  x = keras.layers.PReLU(alpha_initializer='zeros', alpha_regularizer=None, alpha_constraint=None, shared_axes=[1,2])(ini)

  ##Subpixel Construction
  sub_layer_2 = Lambda(lambda x:tf.nn.space_to_depth(x,2)) 
  init = sub_layer_2(inputs=x)


  ##Learning Residual (DCNN)
  ####Conv 3x3x64x64 + PReLu
  x = keras.layers.Conv2D(filters = 64, #feature map number
                     kernel_size = 3, 
                     strides = 1,  # 2
                     padding = 'same', 
                     activation = 'relu',
                     input_shape = (None,None,1))(init)
  x = keras.layers.BatchNormalization()(x)                     
  x = keras.layers.Conv2D(filters = 64, #feature map number
                     kernel_size = 3, 
                     strides = 1,  # 2
                     padding = 'same', 
                     activation = 'relu',
                     input_shape = (None,None,64))(x)
  x = keras.layers.BatchNormalization()(x)                     
  x = keras.layers.Conv2D(filters = 64, #feature map number
                     kernel_size = 3, 
                     strides = 1,  # 2
                     padding = 'same', 
                     activation = 'relu',
                     input_shape = (None,None,64))(x)
  x = keras.layers.BatchNormalization()(x)                     
  # x = keras.layers.MaxPooling2D((2,2),padding='same')(x)
  # x = keras.layers.BatchNormalization()(x)  
  # x = keras.layers.PReLU(alpha_initializer='zeros', alpha_regularizer=None, alpha_constraint=None, shared_axes=[1,2])(x)

  ####Residual Block
  for i in range(6):
    start = x
    Conv1 = keras.layers.Conv2D(filters=64,
                        kernel_size = 1, 
                        strides = 1,  # 2
                        padding = 'same',                                             
                        input_shape = (None,None,64))(start)
    Conv1 = keras.layers.LeakyReLU()(Conv1)
    # Conv1 = keras.layers.MaxPooling2D((2,2),padding='same')(Conv1)
    # Conv1_BN = keras.layers.BatchNormalization()(Conv1)
    # Conv1_BN = Dropout(0.5)(Conv1_BN)
    
    # PReLu = keras.layers.PReLU(alpha_initializer='zeros', alpha_regularizer=None, alpha_constraint=None, shared_axes=[1,2])(Conv1)
    Conv2 = keras.layers.Conv2D(filters=64,
                        kernel_size = 3, 
                        strides = 1,  # 2
                        padding = 'same',                                                
                        input_shape = (None,None,64))(Conv1)
    Conv2 = keras.layers.LeakyReLU()(Conv2)
    # Conv2 = keras.layers.MaxPooling2D((2,2),padding='same')(Conv2)
    # Conv2_BN = keras.layers.BatchNormalization()(Conv2)
    # Conv2_BN = Dropout(0.5)(Conv2_BN)                        
    # PReLu = keras.layers.PReLU(alpha_initializer='zeros', alpha_regularizer=None, alpha_constraint=None, shared_axes=[1,2])(Conv2)
    Conv3 = keras.layers.Conv2D(filters=64,
                        kernel_size = 1, 
                        strides = 1,  # 2
                        padding = 'same',         
                        input_shape = (None,None,64))(Conv2)
    # Conv3 = keras.layers.MaxPooling2D((2,2),padding='same')(Conv3)
    # Conv3 = keras.layers.LeakyReLU()(Conv3)
    
    # Concatenate        
    x = keras.layers.Add()([Conv3,x])
    
  # x = keras.layers.MaxPooling2D((2,2),padding='same')(x)
  ####Conv 3x3x64x64 + PReLu
  x = keras.layers.Conv2D(filters = 64,
                     kernel_size = 3, 
                     strides = 1,  # 2
                     padding = 'same',
                     activation = 'relu',
                     input_shape = (None,None,64))(x)
  x = keras.layers.Conv2D(filters = 64,
                     kernel_size = 3, 
                     strides = 1,  # 2
                     padding = 'same',
                     activation = 'relu',
                     input_shape = (None,None,64))(x)
  x = keras.layers.Conv2D(filters = 64,
                     kernel_size = 3, 
                     strides = 1,  # 2
                     padding = 'same',
                     activation = 'relu',
                     input_shape = (None,None,64))(x)            
  # x = keras.layers.MaxPooling2D((2,2),padding='same')(x)
  # x = keras.layers.BatchNormalization()(x)                                                                 
  # x = keras.layers.PReLU(alpha_initializer='zeros', alpha_regularizer=None, alpha_constraint=None, shared_axes=[1,2])(x)


  ####Conv 3x3x64x48
  x = keras.layers.Conv2D(filters = 64,
                     kernel_size = 3, 
                     strides = 1,  
                     padding = 'same',
                     activation = 'relu',                   
                     input_shape = (None,None,64))(x)
  x = keras.layers.Conv2D(filters = 64,
                     kernel_size = 3, 
                     strides = 1,  
                     padding = 'same',
                     activation = 'relu',                   
                     input_shape = (None,None,64))(x)
  x = keras.layers.Conv2D(filters = 48,
                     kernel_size = 3, 
                     strides = 1,  
                     padding = 'same',
                     activation = 'relu',                   
                     input_shape = (None,None,64))(x)
  # x = keras.layers.MaxPooling2D((2,2),padding='same')(x)                     
  # x = keras.layers.BatchNormalization()(x)

  ###########Learning Residual (DCNN)############
  

  ##Recovery From Subpixel
  sub_layer = Lambda(lambda x:tf.nn.depth_to_space(x,4)) 
  Residual_Output = sub_layer(inputs=x)
  

  ##Initial Prediction
  R = Lambda(lambda x: x[:,:,:,0])(init)
  G = Lambda(lambda x: x[:,:,:,1:3])(init)
  G = Lambda(lambda x: K.mean(x, axis=3))(G)
  B = Lambda(lambda x: x[:,:,:,3])(init)
  # print(init.shape)
  # print(R.shape)
  # print(G.shape)
  # print(B.shape)
  R = Lambda(lambda x: tf.expand_dims(x, -1))(R)
  G = Lambda(lambda x: tf.expand_dims(x, -1))(G)
  B = Lambda(lambda x: tf.expand_dims(x, -1))(B)
  
  #rgb = tf.keras.backend.stack((R, G,B),axis =  3)
  # print(R.shape)
  rg = keras.layers.Concatenate(axis = 3)([R , G])
  rgb = keras.layers.Concatenate(axis = 3)([rg,B])
  # print(rgb.shape)
  Coarse_Output = keras.layers.UpSampling2D(size=(4, 4), interpolation="bilinear")(rgb) #size=4 , from W/2,H/2 ---> 2W,2H
  # Coarse_Output = keras.layers.MaxPooling2D((2,2),padding='same')(Coarse_Output)


  ## + 
  outputs = keras.layers.Add()([Residual_Output,Coarse_Output])
  #outputs = Residual_Output
  model = keras.Model(inputs=inputs, outputs=outputs, name="JDMSR_model")  
  return model

model = create_model()
model.summary()

batch_size = 16
lr = 0.001
e_num = 50
dir_path = 'drive/My Drive/Colab Notebooks/undergraduate_project'

model = create_model()
model.summary()

sgd = SGD(lr=lr, decay=1e-6, momentum=0.9, nesterov=True, clipnorm=1.0)

model.compile(optimizer=Adam(), loss = 'mean_squared_error', metrics = ['mse'])

checkpoint = ModelCheckpoint(os.path.join(dir_path,'model.hdf5'),verbose=1, monitor='loss', save_best_only=True, save_weights_only=True)
rrp = ReduceLROnPlateau(monitor='val_loss', factor=0.1, patience=2, verbose=1, mode='min', min_lr=0.000002)

early_stopping = EarlyStopping(monitor='loss', patience=10, verbose=1, mode='auto')

print(train_image.shape)
print(train_label.shape)
history = model.fit(train_image, train_label, epochs=e_num, batch_size=batch_size,verbose=1,validation_split = 0.1,callbacks=[checkpoint, rrp],shuffle = True)

# Plotting
import matplotlib.pyplot as plt
fig = plt.figure()
plt.plot()
plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
plt.title('Accuracy - 2')
plt.ylabel('accuracy')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='lower right')
plt.savefig('model_accuracy2.png')
plt.show()

plt.plot()
plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.title('Loss - 2')
plt.ylabel('loss')
plt.xlabel('epoch')
plt.legend(['train', 'test'], loc='upper right')
plt.savefig('model_loss2.png')
plt.show()

"""**Validation**"""

from keras.models import load_model
from keras.layers import Lambda
from keras.preprocessing import image
import keras
import tensorflow as tf 
import numpy as np
import matplotlib.pyplot as plt
from PIL import Image 
from keras import backend as K
import os
import math
import shutil

dir_path = 'drive/My Drive/Colab Notebooks/undergraduate_project'
model = create_model()
model.load_weights(os.path.join(dir_path,'model.hdf5'))

input_path = os.path.join(dir_path, 'koda')
output_path = os.path.join(dir_path, 'output')
output_path2 = os.path.join(dir_path, 'output2')
output_path3 = os.path.join(dir_path, 'output3')

if os.path.exists(output_path):
    shutil.rmtree(output_path)
os.makedirs(output_path)

if os.path.exists(output_path2):
    shutil.rmtree(output_path2)
os.makedirs(output_path2)

if os.path.exists(output_path3):
    shutil.rmtree(output_path3)
os.makedirs(output_path3)

entries = os.listdir(input_path)

for entry in entries:
    # Test Image
    print('-----------')
    path = input_path+'/'+entry
    test_image = Image.open(path)
    print(test_image.size)
    
    if (test_image.size[0]/2)%2 != 0: #odd size
      test_image = test_image.resize((test_image.size[0]-1, test_image.size[1]), Image.BICUBIC)
    if (test_image.size[1]/2)%2 != 0:
      test_image = test_image.resize((test_image.size[0], test_image.size[1]-1), Image.BICUBIC)
    path = output_path+'/'+entry
    test_image.save(path)
    print(test_image.size)

    test_image = test_image.resize((test_image.size[0]//2, test_image.size[1]//2), Image.BICUBIC)
    print(test_image.size)
    test_image_array = np.array(test_image)
    test_image_array = bayer_reverse(test_image_array)
    test_image_array = test_image_array[:,:,np.newaxis]
    test_image = image.array_to_img(test_image_array)
    test_image.save(output_path3+'/'+entry)
    # print(test_image.shape)
    print(test_image.size)
    
    
    # print(test_image.shape)
    test_image = np.array(test_image)
    test_image = test_image[np.newaxis,:,:]
    try:
      out = model.predict(test_image)
      out = out[0]
      out = image.array_to_img(out)
      path = output_path2+'/'+entry
      out.save(path)

    except:
      path = input_path+'/'+entry
      test_image = Image.open(path)
      if (test_image.size[0]//scale)%2 != 0:
        test_image = test_image.resize(((test_image.size[0]//scale) - 1, test_image.size[1]//scale), Image.BICUBIC)
      if (test_image.size[1]//scale)%2 != 0:
        test_image = test_image.resize((test_image.size[0]//scale , (test_image.size[1]//scale) - 1), Image.BICUBIC)
      test_image_array = np.array(test_image)
      test_image_array = bayer_reverse(test_image_array)
      test_image_array = test_image_array[:,:,np.newaxis]
      test_image = image.array_to_img(test_image_array)
      print(test_image.size)
      test_image.save(output_path3+'/'+entry)
      test_image = np.array(test_image)
      test_image = test_image[np.newaxis,:,:]
      out = model.predict(test_image)
      out = out[0]
      out = image.array_to_img(out)
      path = output_path2+'/'+entry
      out.save(path)


    # print(out.shape)
    

# print(len(entries))

"""**Performance**"""

import cv2
import numpy as np
import math
import os
from google.colab.patches import cv2_imshow
from skimage.measure import compare_ssim
from skimage.measure import compare_ssim as ssim

def calculate_psnr(original, contrast):
    mse = np.mean((original - contrast) ** 2)
    if mse == 0:
        return 100
    max_value = 255.0
    return 10 * math.log10(max_value**2 / mse)    
    

path = 'drive/My Drive/Colab Notebooks/undergraduate_project'
input_path = os.path.join(path,'output')
output_path = os.path.join(path,'output2')
entries = os.listdir(input_path)
count = 0
total_psnr = 0.
total_ssim = 0.

for entry in entries:  
  img1 = cv2.imread(os.path.join(input_path,entry))
  img2 = cv2.imread(os.path.join(output_path,entry))
  img1 = cv2.resize(img1, (img2.shape[1], img2.shape[0]), interpolation=cv2.INTER_CUBIC)
    


  # PSNR
  psnr = calculate_psnr(img1,img2)
  print("PSNR-{0}: {1:.10f}dB".format(entry,psnr))

  # SSIM
  ssim = compare_ssim(img1, img2, data_range=255.0 - 0.0,multichannel=True)
  print("SSIM-{0}: {1:.10f}".format(entry,ssim))

  total_psnr += psnr
  total_ssim += ssim
  count += 1
print(count)

total_psnr = total_psnr / count
total_ssim = total_ssim / count
print("\n=====================================")
print("Average PSNR:{:.10f}".format(total_psnr))
print("Average SSIM:{:.10f}".format(total_ssim))