Last year I wrote about loading and saving video in a Jupyter Notebook for frame-by-frame processing with PyTorch. Today I’d like to explain more of the actual image processing that motivated me. It started from tinkering with Apple’s Depth Pro model. I just wanted to see how it performed with some arbitrary video and maybe use it to separate the background from the foreground. Today I’ll focus on just two main tasks that differ from the last notebook:

Today’s goal is just to load a video, display individual frames in the output from a Jupyter notebook cell, and write the video back out to a new file. In the middle I’ll do a little processing on the video frames. The processing is beside the point today - I just want to make the input, interaction, and output work really well so that later I can focus more on that processing step in the middle.