tf-video-preprocessing

TensorFlow video preprocessing layers

While working on a video classification model, I was in need of a video augmentation layers, however TensorFlow only provided image processing functions and layers. This led me to the creation of layers for a video processing pipeline.

Most of the implemented functions are analogous to their image counterparts from TensorFlow and Keras.

So far, these are implemeted:

• Random zoom (VideoRandomZoom)
• Random rotation (videoRandomRotation)
• Random crop (VideoRandomCrop)
• Random contrast (VideoRandomContrast)
• Random flip (VideoRandomFlip)
• Resizing
• Center crop
• Rescaling
• Random translation
• Random height
• Random width

Install

pip install tf-video

Usage

Use both using Sequential and Functional TF API, as you would normally do with image preprocessing layers

import tensorflow as tf
from tf_video import VideoRandomFlip, VideoRandomContrast

input_video = ...
model = tf.keras.models.Sequential([
    VideoRandomFlip('horizontal_and_vertical'),
    VideoRandomContrast(0.3)
])

aug = model(input_video)