orphic
  • How it works
  • Catalogue
  • For AIs
  • Open source
SSeeaarrcchh ttoooollss⌘K
orphic

Orpheus went into the underworld to bring back what was lost. Orphic goes into reels, DMs and link-in-bio pages and brings back the real link.

  • How it works
  • Catalogue
  • For AIs
  • Open source
© 2026 Orphic contributors · MIT · made by Adarsh PandeyEvery find credits its creator.
Catalogue/Research learning

Resource

Deepfake detection from your pulse guide (@harixganesh)

Plain-English guide to spotting deepfakes from the pulse visible in skin colour (remote photoplethysmography, as in Intel's FakeCatcher), plus a weekend-sized Python build plan: OpenCV + MediaPipe face landmarks, band-pass filtering and FFT with SciPy/NumPy, patch coherence, a simple classifier, and honest pitfalls (compression, short clips, skin tone, ~91-96% claims).

OOppeenn tthhee ssaavveedd ccooppyyOriginal on docs.google.com ↗
Type
Resource
Category
Research learning
Verified link
Checked
First found
30 Oct 2026

Use it for

  • ↳Learn how pulse-based deepfake detection works
  • ↳Build a heart-rate-from-video reader in python
  • ↳A weekend computer vision project
  • #deepfake-detection
  • #computer-vision
  • #python
  • #learning

From the same post

  • FakeCatcher paperarxiv.org
  • rPPG-Toolboxgithub.com
  • UBFC-rPPG datasetsites.google.com
  • yarPPGsamproell.github.io

More in research learning

  • FakeCatcher paperCiftci, Demir and Yin's paper (Binghamton/Intel) 'FakeCatcher: Detection of Synthetic Portrait Videos using Biological Signals', the basis of Intel's FakeCatcher product.arxiv.org
  • rPPG-ToolboxNeurIPS 2023 open-source toolbox for remote photoplethysmography: run classic (POS, CHROM) and deep models that read heart rate from face video, with benchmarks on datasets like UBFC-rPPG.github.com
  • yarPPGHobby Python project with a deep-dive tutorial that goes from webcam video to heart rate step by step; good for seeing rPPG work quickly.samproell.github.io
  • GPT ResearcherAutonomous research agent that turns one question into a cited report, using any LLM provider and web or local data.github.com