Your first result¶

About five minutes, ending in two video files you can play side by side.

Get the sample clips¶

The clips the original authors used are face.mp4 and baby.mp4 from the MIT project page. From a checkout of this repository:

vidmag download face baby

That needs the repository, not just the installed package: it wraps scripts/download_samples.py, which holds the addresses and is not in the wheel. With only the library installed, fetch the two files yourself into a data/ directory.

Make a pulse visible¶

vidmag magnify data/face.mp4 pulse.mp4 --preset pulse

Play pulse.mp4 next to the original. The face takes on a green flush that appears and fades a little more than once a second — blood arriving with each heartbeat, changing the skin's colour by less than one step in 255, far too little for an eye to catch. The command prints which backend it used before it starts.

Make a small movement visible¶

vidmag magnify data/baby.mp4 breathing.mp4 --preset motion

The sleeping child's chest now visibly rises and falls. The real movement is under a millimetre.

The same thing from Python¶

import vidmag

vidmag.magnify("data/face.mp4", preset="pulse", out="pulse.mp4")

Or on frames you already have, with no file:

import numpy as np
import vidmag

frames = np.zeros((90, 64, 64, 3), dtype=np.uint8)  # your frames, 8-bit BGR
amplified = vidmag.magnify(frames, preset="motion")
print(amplified.shape, amplified.dtype)

If the result looks identical to the input¶

Common, and usually the frequency band. The pulse preset keeps 0.83 to 1.0 cycles per second — a resting heart rate — and a clip shorter than about six seconds at 30 frames per second cannot resolve that band at all. The library warns when that happens and says how many frames it needs. What can go wrong covers the rest.