[BUG] Seed should be passed to image_data_generator.get_random_transform

[Bienqq]

Seed should be passed to this function. In situation when we want to perform the same agumentation for two Image generators this leads to agumentation inconsistency

It should be like this:
params = self.image_data_generator.get_random_transform(x.shape, self.seed)

Instead of
params = self.image_data_generator.get_random_transform(x.shape)

[Dref360]

Hello!

I see that we set the seed here: https://github.com/keras-team/keras-preprocessing/blob/master/keras_preprocessing/image/iterator.py#L58

COuld you share an example reproducing your issue?

[Bienqq]

I use U-Net to perform image segmentation.
I want provide the same seed for image_datagen and mask_datagen to ensure the transformation for image and mask is the same.

def generator(batch_size, train_path, image_folder, mask_folder, aug_dict, image_color_mode="grayscale",
              mask_color_mode="grayscale", image_save_prefix="image", mask_save_prefix="mask", save_to_dir=None,
              target_size=(256, 256), seed=1):

    image_datagen = ImageDataGenerator(**aug_dict)
    mask_datagen = ImageDataGenerator(**aug_dict)

    image_generator = image_datagen.flow_from_directory(
        train_path,
        classes=[image_folder],
        class_mode=None,
        color_mode=image_color_mode,
        target_size=target_size,
        batch_size=batch_size,
        save_to_dir=save_to_dir,
        save_prefix=image_save_prefix,
        seed=seed)

    mask_generator = mask_datagen.flow_from_directory(
        train_path,
        classes=[mask_folder],
        class_mode=None,
        color_mode=mask_color_mode,
        target_size=target_size,
        batch_size=batch_size,
        save_to_dir=save_to_dir,
        save_prefix=mask_save_prefix,
        seed=seed)

    train_generator = zip(image_generator, mask_generator)
    for (img, mask) in train_generator:
        img, mask = adjustData(img, mask)
        yield (img, mask)

[Dref360]

Hello,
I tried to reproduce your issue with the following script. I see no issue with it. Can you provide a self-contained end-to-end example?

import os

import numpy as np
from PIL import Image
from keras.preprocessing.image import ImageDataGenerator

pjoin = os.path.join

cls1_pt = '/tmp/cls1/cls1'
cls2_pt = '/tmp/cls1/cls2'
os.makedirs(cls1_pt, exist_ok=True)
os.makedirs(cls2_pt, exist_ok=True)

for i in range(100):
    img = Image.fromarray((np.ones([100, 100, 3]) * i % 255).astype(np.uint8))
    img.save(pjoin(cls1_pt, '{}.png'.format(i)))
    img.save(pjoin(cls2_pt, '{}.png'.format(i)))

flow1 = ImageDataGenerator(rescale=1 / 255.,
                           rotation_range=10, horizontal_flip=True).flow_from_directory('/tmp/cls1',
                                                                                        class_mode=None,
                                                                                        shuffle=True,
                                                                                        seed=123)
flow2 = ImageDataGenerator(rescale=1 / 255., rotation_range=10,
                           horizontal_flip=True).flow_from_directory('/tmp/cls1', class_mode=None,
                                                                     shuffle=True, seed=123)

gen = zip(flow1, flow2)
for _ in range(10):
    im1, im2 = next(gen)
    assert np.allclose(im1, im2)

[Bienqq]

Your snipped looks and works fine. But I don't know why I am receiving inconsistently transformed images when comparing saved agumented images from both generators

[Keragen]

I had the same issue using Keras for R and an older version of TF. I just found out that using brightness_range in the image_data_generator was causing the problem. using a preprocessing_function and image_modulate from magick to deal with the brightness solved the problem for me.