Question

我正在尝试使用keras裁剪图像数据生成器中的图像中心。我有大小为192x192的图片，我想裁剪它们的中心，以便输出批次为150x150或类似的东西。

我可以立即在Keras ImageDataGenerator中这样做吗？我想不，因为我看到datagenerator中的target_size参数会破坏图像。

我找到了随机裁剪的链接： https://jkjung-avt.github.io/keras-image-cropping/

我已按如下方式修改了作物：

def my_crop(img, random_crop_size):
  if K.image_data_format() == 'channels_last':
    # Note: image_data_format is 'channel_last'
    assert img.shape[2] == 3
    height, width = img.shape[0], img.shape[1]
    dy, dx = random_crop_size #input desired output size
    start_y = (height-dy)//2
    start_x = (width-dx)//2
    return img[start_y:start_y+dy, start_x:(dx+start_x), :]
  else:
      assert img.shape[0] == 3
      height, width = img.shape[1], img.shape[2]
      dy, dx = random_crop_size  # input desired output size
      start_y = (height - dy) // 2
      start_x = (width - dx) // 2
      return img[:,start_y:start_y + dy, start_x:(dx + start_x)]

def crop_generator(batches, crop_length):
    '''
    Take as input a Keras ImageGen (Iterator) and generate
    crops from the image batches generated by the original iterator
    '''
    while True:
        batch_x, batch_y = next(batches)
       #print('the shape of tensor batch_x is:', batch_x.shape)
        #print('the shape of tensor batch_y is:', batch_y.shape)
        if K.image_data_format() == 'channels_last':
         batch_crops = np.zeros((batch_x.shape[0], crop_length, crop_length, 3))
        else:
         batch_crops = np.zeros((batch_x.shape[0], 3, crop_length, crop_length))
        for i in range(batch_x.shape[0]):
            batch_crops[i] = my_crop(batch_x[i], (crop_length, crop_length))
        yield (batch_crops, batch_y)

这个解决方案对我来说似乎很慢，请问还有其他更有效的方法吗？你会建议什么？

先谢谢

Answer 1

我试图用这种方式解决它：

offset

如果你有更好的方法，请提出你的建议。

Answer 2

您可以像这样定义自定义裁剪功能。

from keras.preprocessing.image import load_img
from keras.preprocessing.image import load_img
from keras.preprocessing.image import img_to_array
from keras.preprocessing.image import array_to_img

def crop(image):
    
    start_y = (img_height - crop_length) // 2
    start_y = (img_width - crop_length) // 2
    cropped_image=image[start_x:(img_width - start_x), start_y:(img_height - start_y), :]
    
    return cropped_image

该函数可以像这样包含在 Keras 自定义生成器中。

ImageDataGenerator(preprocessing_function=crop)

然后你可以根据你的选择ImageDataGenerator对象

在Keras ImageDataGenerator或flow_from_directory中裁剪图像的中心

2 个答案: