ONLINEREV · 2026-05-13§ RESEARCH · MOTION-MAGNIFICATIONPLATFORM · iPad / MetalPUB · 2026-03-26
All Research
§ Case StudyiPad / Metal GPU

Motion Magnification

From MIT research papers to real-time GPU processing on iPad — using AI to reverse-engineer published signal processing algorithms into custom Apple Metal compute shaders.

Computer VisionGPU ComputingSignal ProcessingMetal ShadersiOS
§ 01

The Science

Motion magnification is a computational technique that amplifies subtle, invisible motions in video to make them visible to the naked eye. A person's face changes color slightly with each heartbeat. A bridge sways imperceptibly in the wind. A machine vibrates at frequencies too small to see. Motion magnification reveals all of it.

This project implements two complementary methods developed by researchers at MIT CSAIL:

§ Method A

Eulerian Video Magnification

Decomposes video into spatial frequency bands using a Laplacian pyramid, applies temporal bandpass filtering to isolate motions of interest, then amplifies and reconstructs. Published at SIGGRAPH 2012 by Wu et al.

§ Method B

Phase-Based Motion Processing

Uses a complex steerable pyramid to decompose video into local phase signals, enabling approximately 4× larger magnification factors with fewer artifacts. Published at SIGGRAPH 2013 by Wadhwa et al.

For the full technical details, see the original MIT CSAIL research: Eulerian Video Magnification and Phase-Based Video Motion Processing.

§ 02

See It in Action

Real-time motion magnification running on an iPad — amplifying subtle wave (and other) motions.

§ 03

The Approach

The goal was to take algorithms described in academic research papers and get them running in real time on consumer mobile hardware — specifically, an iPad. This required translating dense mathematical notation into high-performance GPU code.

Using Claude as an AI research assistant, the original MIT papers were analyzed to extract the core signal processing and image processing algorithms. Claude helped reverse-engineer the mathematical foundations — Laplacian pyramids, complex steerable pyramids, IIR and biquad temporal filters, DCT-based ideal bandpass filters — and translate them into custom Apple Metal compute shaders (GPU kernels) optimized for mobile hardware.

The result is multiple real-time processing pipelines running entirely on the GPU:

§ Pipeline A · Eulerian (IIR)~30 fps

Real-time live camera processing with first-order IIR temporal bandpass filtering across a 5-level Laplacian pyramid.

§ Pipeline B · Phase-Based~17–20 fps

Real-time phase extraction and temporal filtering using a complex steerable pyramid, powered by custom Metal GPU compute shaders.

The application also supports batch processing modes with higher-quality DCT ideal bandpass filtering, FLIR thermal camera integration, and interactive region-of-interest selection for targeted magnification.

§ 04

Interested in This Technology?

The Eulerian and phase-based motion magnification methods are covered by US Patent 9,811,901 B2, jointly owned by the Massachusetts Institute of Technology and Quanta Computer Inc. (active through January 2033). As a result, this software is not available for sale.

However, Slick Engineering has a fully working, real-time implementation of both methods running on consumer mobile hardware. If you or your organization are interested in licensing the patent from MIT and Quanta Computer and would like a development partner who already has working code, we would welcome the conversation.