Mohammad Rami Koujan

Mohammad Rami Koujan

Manager, Machine Learning Engineering

Generative ML Platform · Snap Inc. · London, UK

CV / Email / Scholar / LinkedIn

I am a Manager of Machine Learning Engineering at Snap Inc. in London, leading a team building on-device generative AI for human-centric AR. Our work ships to the hundreds of millions of people on Snapchat (953M+ MAU) and to creators on Lens Studio Pro and EasyLens.

My research and engineering focus on controllable diffusion models for high-fidelity human image and video generation, distillation of large image and video diffusion models into real-time, lightweight lenses on mobile GPUs/NPUs, and the broader stack of neural avatars, virtual try-on, and 3D face/body reconstruction. I have shipped several generative AR features at Snap and am co-inventor on a number of US and UK patents in real-time face and body synthesis.

I hold a PhD in Computer Vision and Machine Learning from the University of Exeter (2021) and an MSc in Computer Vision and Robotics (VIBOT, Erasmus Mundus, with Distinction). Previously I worked at Ariel AI (acquired by Snap in 2020), at Facesoft — an Imperial College London spin-out whose technology assets were acquired by Huawei — and as a visiting researcher in Imperial's iBUG group with Prof. Stefanos Zafeiriou. I have peer-reviewed publications at CVPR, FG, IJCB, and SIGGRAPH (371 Google Scholar citations as of May 2026).

If you'd like to chat about generative AI, AR, or hiring, reach out at mhd.kou@gmail.com.

News

May 2026US patents 18/429,251 (Automatic Image Generation Using Latent Structural Diffusion) and 18/226,929 (Generating Ground Truths for Generative AI Applications) granted.
Apr 2026US patent 12,602,850 granted: Generative AI Virtual Clothing Try-On.
Dec 2025Character Skin Generator shipped in Lens Studio Pro and as Body Avatar in EasyLens — a flagship real-time on-device avatar feature.
Oct 2025US patents 12,499,638 (Stylizing a Whole-body of a Person) and 12,437,367 (Real-time Try-on Using Body Landmarks) granted.
Jul 2025Promoted to Manager, Machine Learning Engineering — leading a GenAI team at Snap.
Feb 2024UK patent GB 2594970 granted: Three-dimensional Facial Motion Estimation.
Jan 2024Promoted to Senior Machine Learning Research Engineer at Snap.
Aug 2022Automatic Stress Analysis from Facial Videos based on Deep Facial Action Units Recognition published in Pattern Analysis and Applications (Springer).
see older news
2021Head2Head++: Deep Facial Attributes Re-targeting published at IEEE IJCB 2021.
Mar 2021PhD awarded — University of Exeter. Thesis: 3D Face Modelling, Analysis and Synthesis.
2020Four papers published at IEEE FG 2020 (Head2Head, ReenactNet, Real-Time Facial Expression Recognition, Automatic Stress Detection).
2020DeepFaceFlow: In-the-Wild Dense 3D Facial Motion Estimation published at IEEE/CVF CVPR 2020.
Sep 2020Joined Snap Inc. through the acquisition of Ariel AI.
Dec 2018Combining Dense Non-Rigid Structure-from-Motion and 3D Morphable Models for Monocular 4D Face Reconstruction published at ACM SIGGRAPH 2018.
Jul 2018Texture Descriptors for Classifying Sparse, Irregularly Sampled Optical Endomicroscopy Images published at the Annual Conference on Medical Image Understanding and Analysis (MIUA).
Apr 2018Multi-class Classification of Pulmonary Endomicroscopic Images published at the 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018).

Selected Publications

371 Google Scholar citations as of May 2026. Authors: own name in bold.

Automatic Stress Analysis: face detection & landmarking + 2D image registration Automatic Stress Analysis from Facial Videos based on Deep Facial Action Units Recognition.
Giannakakis, Mohammad Rami Koujan, Roussos.
Pattern Analysis and Applications, Springer, 2022.
[ paper ]
[ bibtex ]
@article{giannakakis2022stress,
  title={Automatic stress analysis from facial videos based on deep facial action units recognition},
  author={Giannakakis, Giorgos and Koujan, Mohammad Rami and Roussos, Anastasios and Marias, Kostas},
  journal={Pattern Analysis and Applications},
  year={2022},
  publisher={Springer}
}
Head2Head++: Turnbull-to-Biden head reenactment Head2Head++: Deep Facial Attributes Re-targeting.
Mohammad Rami Koujan, Doukas, Sharmanska, Roussos, Zafeiriou.
IEEE IJCB 2021.
[ paper ] [ arxiv ] [ demo ] [ code ]
[ bibtex ]
@inproceedings{koujan2021head2headpp,
  title={Head2Head++: Deep Facial Attributes Re-targeting},
  author={Koujan, Mohammad Rami and Doukas, Michail Christos and Sharmanska, Viktoriia and Roussos, Anastasios and Zafeiriou, Stefanos},
  booktitle={IEEE International Joint Conference on Biometrics (IJCB)},
  year={2021}
}
DeepFaceFlow: aligned facial image and 3D flow visualization DeepFaceFlow: In-the-Wild Dense 3D Facial Motion Estimation.
Mohammad Rami Koujan, Roussos, Zafeiriou.
IEEE/CVF CVPR 2020.
[ paper ] [ supp ] [ arxiv ] [ demo ] [ slides ]
[ bibtex ]
@inproceedings{koujan2020deepfaceflow,
  title={DeepFaceFlow: In-the-Wild Dense 3D Facial Motion Estimation},
  author={Koujan, Mohammad Rami and Roussos, Anastasios and Zafeiriou, Stefanos},
  booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2020}
}
Head2Head: full head reenactment between two subjects Head2Head: Video-based Neural Head Synthesis.
Mohammad Rami Koujan, Doukas, Roussos.
IEEE FG 2020.
[ paper ] [ arxiv ] [ demo ] [ code ] [ slides ] [ talk ]
[ bibtex ]
@inproceedings{koujan2020head2head,
  title={Head2Head: Video-based Neural Head Synthesis},
  author={Koujan, Mohammad Rami and Doukas, Michail Christos and Roussos, Anastasios},
  booktitle={IEEE International Conference on Automatic Face and Gesture Recognition (FG)},
  year={2020}
}
ReenactNet: source actor, 3D mesh, and reenacted target ReenactNet: Real-time Full Head Reenactment.
Mohammad Rami Koujan, Doukas, Roussos, Zafeiriou.
IEEE FG 2020.
[ paper ] [ arxiv ] [ demo ]
[ bibtex ]
@inproceedings{koujan2020reenactnet,
  title={ReenactNet: Real-time Full Head Reenactment},
  author={Koujan, Mohammad Rami and Doukas, Michail Christos and Roussos, Anastasios and Zafeiriou, Stefanos},
  booktitle={IEEE International Conference on Automatic Face and Gesture Recognition (FG)},
  year={2020}
}
Real-Time Facial Expression Recognition In the Wild Real-Time Facial Expression Recognition "In the Wild" by Disentangling 3D Expression from Identity.
Mohammad Rami Koujan, Alharbawee, Giannakakis, Pugeault, Roussos.
IEEE FG 2020.
[ paper ] [ arxiv ] [ demo ] [ slides ] [ talk ]
[ bibtex ]
@inproceedings{koujan2020realtime,
  title={Real-Time Facial Expression Recognition "In the Wild" by Disentangling 3D Expression from Identity},
  author={Koujan, Mohammad Rami and Alharbawee, Luma and Giannakakis, Giorgos and Pugeault, Nicolas and Roussos, Anastasios},
  booktitle={IEEE International Conference on Automatic Face and Gesture Recognition (FG)},
  year={2020}
}
Stress Detection title slide: Automatic stress detection evaluating models of facial action units Automatic Stress Detection Evaluating Models of Facial Action Units.
Giannakakis, Mohammad Rami Koujan, Roussos, Marias.
IEEE FG 2020.
[ paper ] [ slides ]
[ bibtex ]
@inproceedings{giannakakis2020stress,
  title={Automatic Stress Detection Evaluating Models of Facial Action Units},
  author={Giannakakis, Giorgos and Koujan, Mohammad Rami and Roussos, Anastasios and Marias, Kostas},
  booktitle={IEEE International Conference on Automatic Face and Gesture Recognition (FG)},
  year={2020}
}
Animated 4D face reconstructions from monocular video Combining Dense Non-Rigid Structure-from-Motion and 3D Morphable Models for Monocular 4D Face Reconstruction.
Mohammad Rami Koujan, Roussos.
ACM SIGGRAPH 2018.
[ project ] [ paper ] [ demo ] [ slides ]
[ bibtex ]
@inproceedings{koujan2018combining,
  title={Combining Dense Non-Rigid Structure-from-Motion and 3D Morphable Models for Monocular 4D Face Reconstruction},
  author={Koujan, Mohammad Rami and Roussos, Anastasios},
  booktitle={ACM SIGGRAPH},
  year={2018}
}
PhD Thesis title slide: 3D Face Modelling, Analysis and Synthesis 3D Face Modelling, Analysis and Synthesis.
Mohammad Rami Koujan.
PhD Thesis, University of Exeter, 2021.
[ thesis ] [ slides ]

Patents

Additional patents pending in real-time on-device generative AI for human image/video synthesis and AR lenses.

Selected Projects

Character Skin Generator

Real-time on-device generative AI that transforms users into stylised digital characters in AR. Architected and shipped the core ML pipeline in Lens Studio Pro.

Shipped · Lens Studio Pro

EasyLens — Body Avatar

Consumer-facing version of Character Skin Generator deployed as a building block in Snap's no-code lens platform, EasyLens.

Shipped · Snapchat / EasyLens

Garment Transfer (Virtual Try-On)

Optimised on-device virtual try-on for upper garments, deployed natively in the Snapchat app. Co-invented; protected by US patent 12,437,367.

Shipped · Snapchat / Lens Studio

Experience

Sep 2020 – PresentSnap Inc., London — Manager, Machine Learning Engineering (Jul 2025 – Present); previously Senior MLRE (Jan 2024 – Jun 2025) and MLRE (Sep 2020 – Dec 2023). Generative ML Platform.
Jul – Sep 2020Ariel AI, London — Computer Vision Scientist. Joined Snap through company acquisition.
Jul 2019 – May 2020Facesoft (Imperial College London spin-out), London — Computer Vision Scientist. Technology asset acquired by Huawei.
Jun 2018 – Aug 2019Imperial College London — iBUG (Prof. Stefanos Zafeiriou) — Visiting Researcher.
Jun 2017 – Jun 2019University of Exeter — Research & Teaching Assistant.
Jan – Jun 2017EPSRC Proteus IRC, Edinburgh — Research Assistant.
Jun – Sep 2016Inria, Nancy — Research Assistant.

Full chronology in the CV.

Education

2017 – 2021 PhD, Computer Vision and Machine Learning — University of Exeter, UK.
Thesis: 3D Face Modelling, Analysis and Synthesis. College-funded studentship. ICVSS 2019 · BMVA Summer School 2018.
2015 – 2017 MSc, Computer Vision and Robotics (VIBOT — Erasmus Mundus) — Heriot-Watt University, University of Girona & Université de Bourgogne.
Graduated with Distinction · 50 K€ Erasmus Mundus Scholarship · Bourgogne Region International Mobility Grant.
2009 – 2014 BSc, Communications and Information Engineering — Yarmouk Private University, Syria.
GPA 3.92/4.0 · Faculty Top Student.

Awards & Honors

Languages. English (full professional) · Arabic (native) · French (limited working) · Catalan (elementary).