How To Do Keras Image Augmentation Using Custom Data Generator?

I am using Keras custom generator and i want to apply image augmentation techniques on data returned from custom data generator. I want these image augmentation techniques ImageDat

Solution 1:

Haven't tried it but I guess you can use the flow method from your instance of ImageDataGenerator. For example, your custom class could look like this:

class CustomDataGenerator(tf.keras.utils.Sequence):
    
    def __init__(self, batch_size=32):
        self.batch_size = batch_size
        self.augmentor = ImageDataGenerator(
            rotation_range=40,
            width_shift_range=0.2,
            height_shift_range=0.2,
            shear_range=0.2,
            zoom_range=0.2,
            horizontal_flip=True,
            fill_mode='nearest'
        )

    ...

    def __data_generation(self, list_IDs_temp):
      'Generates data containing batch_size samples' # X : (n_samples, *dim, n_channels)
      # Initialization
      X = np.empty((self.batch_size, *self.dim, self.n_channels))
      y = np.empty((self.batch_size), dtype=int)

      # Generate data
      for i, ID in enumerate(list_IDs_temp):
          # Store sample
          X[i,] = tfk.preprocessing.image.load_img(self.list_IDs[ID])
    
          # Store class
          y[i] = self.labels[ID]

      X_gen = self.augmentor.flow(X, batch_size=self.batch_size, shuffle=False)
      """do not perform shuffle here, the shuffling is performed beforehand
       by your custom class anyway, you just want the transformations to be 
      applied, and above all you want to keep your images synced with the 
      labels.""" 
      
      return next(X_gen), tkf.utils.to_categorical(y, num_classes=self.n_classes)