TensorFlow代码

数据处理：

输入为（344,288）的图片，首先进行归一化处理，然后下采样再上采样，得到输入input_，

• 
  
  def preprocess(path, scale=3):
  
  
• 
  
    """
  
  
• 
  
    Preprocess single image file
  
  
• 
  
      (1) Read original image as YCbCr format (and grayscale as default)
  
  
• 
  
      (2) Normalize
  
  
• 
  
      (3) Apply image file with bicubic interpolation
  
  
• 
  
  
  
  
• 
  
    Args:
  
  
• 
  
      path: file path of desired file
  
  
• 
  
      input_: image applied bicubic interpolation (low-resolution)
  
  
• 
  
      label_: image with original resolution (high-resolution)
  
  
• 
  
    """
  
  
• 
  
    image = imread(path, is_grayscale=True)#(344,228)
  
  
• 
  
    label_ = modcrop(image, scale)#(342,228)modcrop得到scale的整数倍
  
  
• 
  
  
  
  
• 
  
    # Must be normalized
  
  
• 
  
    image = image / 255.
  
  
• 
  
    label_ = label_ / 255.
  
  
• 
  
  
  
  
• 
  
    input_ = scipy.ndimage.interpolation.zoom(label_, (1./scale), prefilter=False)#(114,76)
  
  
• 
  
    input_ = scipy.ndimage.interpolation.zoom(input_, (scale/1.), prefilter=False)#(342,228)
  
  
• 
  
    return input_, label_
  
  
  

得到子图片

• 
  
  input_, label_ = preprocess(data[0], config.scale)
  
  
• 
  
  
  
  
• 
  
      if len(input_.shape) == 3:
  
  
• 
  
        h, w, _ = input_.shape
  
  
• 
  
      else:
  
  
• 
  
        h, w = input_.shape#(342,228)
  
  
• 
  
  
  
  
• 
  
      # Numbers of sub-images in height and width of image are needed to compute merge operation.
  
  
• 
  
      nx = ny = 0
  
  
• 
  
      for x in range(0, h-config.image_size+1, config.stride):#(0,310,21)nx=15,ny=10
  
  
• 
  
        nx += 1; ny = 0
  
  
• 
  
        for y in range(0, w-config.image_size+1, config.stride):#(0,196,21)
  
  
• 
  
          ny += 1
  
  
• 
  
          sub_input = input_[x:x+config.image_size, y:y+config.image_size] # [33 x 33]
  
  
• 
  
          sub_label = label_[x+int(padding):x+int(padding)+config.label_size, y+int(padding):y+int(padding)+config.label_size] # [21 x 21]
  
  
• 
  
  
  
  
• 
  
          sub_input = sub_input.reshape([config.image_size, config.image_size, 1])  #(33,33,1)
  
  
• 
  
          sub_label = sub_label.reshape([config.label_size, config.label_size, 1])  #(21,21,1)
  
  
• 
  
  
  
  
• 
  
          sub_input_sequence.append(sub_input)
  
  
• 
  
          sub_label_sequence.append(sub_label)
  
  
• 
  
  
  
  
• 
  
    """
  
  
• 
  
    len(sub_input_sequence) : the number of sub_input (33 x 33 x ch) in one image
  
  
• 
  
    (sub_input_sequence[0]).shape : (33, 33, 1)
  
  
• 
  
    """
  
  
• 
  
    # Make list to numpy array. With this transform
  
  
• 
  
    arrdata = np.asarray(sub_input_sequence) # [?, 33, 33, 1]
  
  
• 
  
    arrlabel = np.asarray(sub_label_sequence) # [?, 21, 21, 1]
  
  
• 
  
  
  
  
• 
  
    make_data(sess, arrdata, arrlabel)
  
  
• 
  
  
  
  
• 
  
    if not config.is_train:
  
  
• 
  
      return nx, ny        #15,10
  
  
  

奈斯，很赞