How to apply Video augmentation with keras preprocessing layers uniformly for all frames in the video?

Question

I'm trying to apply data augmentation to a video dataset wherein each video is applied with the different augmentations. For example, all frames in video 1 are flipped horizontally and rotated by 10°. All frames in video 2 on the other hand, are not flipped and rotated by -5°. I passed a seed in the preprocessing layers, however, each frames of video 1 are augmented differently. This is what my approach looks:

def data_augment(frames,seed):
    x = tf.keras.layers.CenterCrop(height=1000,width=1200) (frames)
    x = Resizing(width=128,height=128) (x)
    x = Rescaling(1./255) (x)
    x = RandomContrast((0.2,0.2),seed=seed) (x)
    x = RandomTranslation(height_factor=0.15,width_factor=0.2,fill_mode="constant",fill_value=0.0,seed=seed) (x)
    x = RandomFlip("horizontal",seed=seed) (x)
    x = RandomRotation(factor=0.01,fill_mode="constant",seed=seed) (x)
    return x

Answer 1

Video Augmentation:: For videos, combine time and channel axis and treat it as an image augmentation problem. And reshape the end result to get videos augmented same for all frames.

#input dimension:
BATCH, TIME,WIDTH, HEIGHT,_= tf.shape(videos)

Step1: change input shape-(batch, time, width, height, 3) to (batch, width, height, time*3)

#move time to last
videos = tf.transpose(videos, [0, 2, 3, 4, 1])

#combine channels and time
out_shape = (BATCH, WIDTH, HEIGHT, TIME*3)

videos = tf.reshape(videos, out_shape)    

Step2: apply augmentation

augmented_data = data_augment(videos,...)

Step 3: reshape back to original

BATCH, WIDTH, HEIGHT,channels= tf.shape(augmented_data)

augmented_data = tf.reshape(augmented_data, (BATCH, HEIGHT, WIDTH, 3, channels//3))
augmented_data = tf.transpose(augmented_data, [0, 4, 1, 2, 3])