pyevm

Eulerian Video Magnification — reveal invisible motion and colour changes in video.

Eulerian Video Magnification (EVM) is a computational technique that amplifies subtle, otherwise invisible variations in video — such as the colour flush of a heartbeat, the micro-vibrations of a bridge, or the barely perceptible breathing of a sleeping animal. pyevm provides clean, GPU-accelerated Python implementations of the three canonical EVM algorithms, plus a CLI and an interactive Streamlit app.

Showcase

Original	Colour magnification	Motion magnification

Algorithms

Installation

pip

pip install pyeulervid

uv

uv add pyeulervid

conda / pixi (conda-forge)

conda install -c conda-forge pyevm
pixi add pyevm

Optional: GPU-accelerated video I/O

torchcodec enables faster video decoding via FFmpeg. pip wheels are Linux-only; on Windows install via pixi (conda-forge provides Windows CUDA builds):

# pip / uv (Linux only)
pip install "pyeulervid[fast-io]"
uv add "pyeulervid[fast-io]"

# conda-forge — works on Windows + Linux + macOS
conda install -c conda-forge pyevm  # torchcodec is included automatically
pixi add pyevm

Quick start

Python API

from pyevm import ColorMagnifier, MotionMagnifier, PhaseMagnifier
from pyevm.io.video import VideoReader, VideoWriter

reader = VideoReader("input.mp4")

# --- Batch mode (loads all frames into memory) ---
frames, fps = reader.read()

magnifier = ColorMagnifier(alpha=50, freq_low=0.4, freq_high=3.0)
result = magnifier.process(frames, fps)

VideoWriter("output.mp4", fps=fps).write(result)

# --- Streaming mode (O(1) memory — recommended for long/HD video) ---
magnifier = MotionMagnifier(alpha=20, freq_low=0.4, freq_high=3.0)

frame_stream, fps, n_frames = reader.stream()
output_stream = magnifier.process_stream(frame_stream, fps, n_frames=n_frames)
VideoWriter("output.mp4", fps=fps).write_stream(output_stream)

CLI

# Colour magnification
pyevm color input.mp4 output.mp4 --alpha 50 --freq-low 0.4 --freq-high 3.0

# Motion magnification
pyevm motion input.mp4 output.mp4 --alpha 20 --freq-low 0.4 --freq-high 3.0

# Phase-based magnification
pyevm phase input.mp4 output.mp4 --factor 10 --freq-low 0.4 --freq-high 3.0

# Inspect detected compute device
pyevm info

# Launch the interactive dashboard
pyevm dashboard

Add --debug to any command for verbose logging, including per-chunk timing breakdowns (pyramid build / filter / collapse) to identify performance bottlenecks.

Streamlit dashboard

pyevm dashboard

CLI reference

pyevm color

Amplifies subtle colour changes (e.g. skin-tone flush from pulse).

pyevm motion

Amplifies subtle physical motion (e.g. breathing, structural vibration).

pyevm phase

Artifact-free motion magnification via steerable pyramid phase decomposition.

pyevm dashboard

Opens the interactive Streamlit dashboard in a browser. Requires streamlit to be installed (pip install streamlit).

pyevm info

Prints detected compute device, PyTorch version, and GPU details.

Hardware

pyevm automatically selects the best available compute device:

1. CUDA — NVIDIA GPU
2. MPS — Apple Silicon GPU
3. CPU — fallback

Override with --device cuda, --device mps, or --device cpu.

References

• Wu, H.-Y., Rubinstein, M., Shih, E., Guttag, J., Durand, F., & Freeman, W. T. (2012). Eulerian Video Magnification for Revealing Subtle Changes in the World. ACM Transactions on Graphics, 31(4). PDF
• Wadhwa, N., Rubinstein, M., Durand, F., & Freeman, W. T. (2013). Phase-Based Video Motion Processing. ACM Transactions on Graphics, 32(4). PDF

Contributing

Bug reports and pull requests are welcome. Please open an issue first for any non-trivial changes.

License

MIT