Showing custom image data in tensorboard by using tf.keras.callbacks.tensorboard only

Question

I am using keras ImageDataGenerator to load my images

train_batches = ImageDataGenerator().flow_from_directory(train_path, target_size=(img_height,img_width), batch_size=16,shuffle=True,class_mode='categorical')
valid_batches = ImageDataGenerator().flow_from_directory(valid_path, target_size=(img_height,img_width), batch_size=8,shuffle=True,class_mode='categorical')

and I use callback function

tensorboard_callback = TensorBoard(
    log_dir=log_dir,
    histogram_freq=1,
    write_graph=True,
    write_images=True,
    write_steps_per_second=True,
    update_freq='epoch',
)

there is no image data when I use callback function by default and if I use callback function from above then the output is not what I want(picture below)

I want my output image result like this URL: How to display custom images in TensorBoard using Keras?

1. tf.summary.image is quite complicate for me. I don't know how to load image from ImageDataGenerator by using this function
2. Can I only using callback fuction to solve my question?