Head of Multimedia AI & Staff Research Engineer

Samsung Electronics | AI & HPC Specialist

15+ years of experience in Artificial Intelligence and HPC research and development as a technical leader and individual contributor. Throughout my career, I have delivered cutting-edge technology solutions across various industries while leading world-class teams. My work has resulted in numerous commercialized solutions, with a strong record of patent filings and publications in reputable conferences. My research interests include: personalized learning, generative AI, augmented reality (AR), parallel computing and computer vision systems. I combine technical expertise with leadership to drive innovation from research to real-world applications.

Get in Touch View Research
Albert Saà Garriga - Head of Multimedia AI

Research Engineer & Technical Leader

I bridge the chasm between theoretical academia and commercialized, on-device technology. My unique expertise combines deep technical mastery in parallel computing and computer vision with high-level executive leadership managing large-scale, distributed research teams.

As Head of Multimedia AI at Samsung Electronics, I lead a team of researchers while remaining deeply involved in hands-on technical work. My role combines technical leadership with strategic direction, including project proposals, patent writing, execution mapping, and MLOps platform development. I continue to contribute as a staff engineer on key projects, ensuring that our research translates into real-world applications.

20+ Patents

Image Processing & AI Techniques

10+ Publications

Top-tier Conferences (CVPR, ICASSP, etc.)

Heterogenous Research Group

Leading Distributed Team Researchers+Enginners

The Visual Timeline

A journey through technical evolution and leadership

2024–Present

Head of Multimedia AI. Staff Research Engineer

Samsung Electronics

Leading group of researchers on GenAI & Omni-LLMs agents.

Gaussian Splatting (GSta, Trick-gs) Neural Rendering (CheapNVS, Patented) Live Effect(Commercialization, Patented)
2021–2023

Head of Multimedia AI. Principal Research Engineer

Samsung Electronics

Focus on commercialization of flagship features.

Semi-supervised object segmentation(MobileVOS, TRICKVOS, Patented) Shadow/Object Eraser(LP-IOANET, Patented) Monocular Depth Estimation(Real-time monodepth, Bokeh) Inverse Rendering(Patented) Speaker Verification(Patented) Federated Learning(Patented)
2016–2020

Project Lead/Individial Contibutor. Senior Research Engineer

Samsung Electronics

Core computer vision R&D.

SLAM(Part of LPGPU European Project) Photo Realistic Style Transfer(Patented) Portrait Relighting(Patented) Makeup Transfer(Patented) Personalized Image Enhancement(Patented) 360 Video Stabilization(Commercialization)
2012–2016

PhD Researcher

CAIAC

HPC & Heterogeneous Computing.

Source Code Adaptation CUDA/MPI Optimization
2010–2013

Junior Developer

Daxip

Early innovation and mobile prototyping.

Augmented Sound Reality TV Ads Detector

Research Pillars

Core areas of technical expertise

Computational Photography

Intelligent restoration and enhancement of visual content for mobile devices.

Shadow Removal: LP-IOANET (ICASSP 2023) - Efficient high resolution document shadow removal

Object Erasure: Shadow/Object eraser (2020) - Magic Eraser style features with mask-based pyramid networks

Relighting: Portrait relighting through inverse rendering techniques

LP-IOANET document shadow removal results

3D Vision & Generative AI

Neural rendering and spatial computing for immersive experiences.

GSta: Efficient training scheme with siestaed gaussians for monocular 3D scene reconstruction (In Submission)

CheapNVS: Real-time on-device narrow-baseline novel view synthesis (ICASSP 2025)

Trick-gs: A balanced bag of tricks for efficient gaussian splatting (ICASSP 2025)

NeRF Exploration: Finding Waldo in 3D space

Trick-gs 3D point clouds and novel views

Video Understanding

Temporal consistency and real-time video processing.

MobileVOS: Video object segmentation using knowledge distillation (CVPR 2023)

TRICKVOS: Contrastive learning meets knowledge distillation (2023)

Applications: Video bokeh, color point effects

MobileVOS video segmentation results

Systems & Privacy

MLOps infrastructure and privacy-preserving AI.

Federated Learning: 5+ patents on FL applied to computer vision

MLOps: DevOps, CI/CD, Docker, GitHub Actions implementation

Privacy: Training models without exchanging user data

Federated learning architecture diagram

Selected Publications & Patents

The visual gallery of research output

GSta: Efficient Training Scheme with Siestaed Gaussians

GSta: Efficient Training Scheme with Siestaed Gaussians for Monocular 3D Scene Reconstruction

In Submission. Anil Armagan, Albert Saà Garriga, Bruno Manganelli, Kyuwon Kim, M. Kerim Yucel. Samsung R&D Institute UK.

View on arXiv

Trick-gs paper title block and architecture diagram

Trick-gs: A balanced bag of tricks for efficient gaussian splatting

Optimizing Gaussian Splatting for mobile architectures. Presented at ICASSP 2025.

View on arXiv

CheapNVS paper header and architecture diagram

CheapNVS: Real-time On-Device Narrow-Baseline Novel View Synthesis

Redefining mobile rendering: Real-time novel view synthesis on edge devices. Presented at ICASSP 2025.

View on arXiv

LP-IOANET document shadow removal visual comparison

LP-IOANET: Efficient High Resolution Document Shadow Removal

Enterprise-grade document digitization via deep learning. Presented at ICASSP 2023.

View on arXiv

MobileVOS architecture diagram

MobileVOS: Real-time Video Object Segmentation

Real-time Video Object Segmentation using Knowledge Distillation. CVPR 2023.

View on arXiv

TRICKVOS architecture diagram

TRICKVOS: A Bag of Tricks for Video Object Segmentation

Contrastive learning meets knowledge distillation. 2023.

View on arXiv

Finding Waldo NeRF exploration

Finding waldo: Towards efficient exploration of nerf scene spaces

Efficient exploration of NeRF scene spaces. 2024.

View on arXiv

Rethinking Encoder-Decoder Flow

Rethinking Encoder-Decoder Flow Through Shared Structures

Shared structures for encoder-decoder flow. ICASSP 2025.

View on arXiv

Real-time Monocular Depth Estimation

Real-time Monocular Depth Estimation with Sparse Supervision on Mobile

Efficient depth estimation for mobile devices with sparse supervision. CVPRW 2020.

View Paper

Adaptive Mask-based Pyramid Network

Adaptive Mask-based Pyramid Network for Realistic Bokeh Rendering

Realistic bokeh rendering using adaptive mask-based pyramid networks. ICASSP 2023.

View on arXiv

Patent front pages collage

Intellectual Property Portfolio

20+ Patents secured in Image Understanding, Device Personalization, and Federated Learning.

View on EspaceNet View on UK Patent Office

Samsung Research Blog Posts

Explore our latest research innovations and breakthroughs

Trick-GS: A Balanced Bag of Tricks for Efficient Gaussian Splatting

Trick-GS: A Balanced Bag of Tricks for Efficient Gaussian Splatting

Discover our optimization techniques for efficient Gaussian Splatting, enabling real-time 3D scene reconstruction on mobile devices.

Read on Samsung Research

CheapNVS: Real-Time On-Device Novel View Synthesis

CheapNVS: Real-Time On-Device Novel View Synthesis

Learn about our approach to real-time novel view synthesis that runs efficiently on mobile devices, enabling immersive experiences.

Read on Samsung Research

Finding Waldo: Towards Efficient Exploration of NeRF Scene Spaces

Finding Waldo: Towards Efficient Exploration of NeRF Scene Spaces

Explore our innovative approach to efficiently navigate and explore NeRF scene spaces, making 3D scene understanding more accessible.

Read on Samsung Research

LP-IOANet: Efficient High Resolution Document Shadow Removal

LP-IOANet: Efficient High Resolution Document Shadow Removal

Dive into our document shadow removal technique that enables high-quality digitization of documents with complex lighting conditions.

Read on Samsung Research

MobileVOS: Real-Time Video Object Segmentation

MobileVOS: Real-Time Video Object Segmentation

Discover how we combine contrastive learning with knowledge distillation to achieve real-time video object segmentation on mobile devices.

Read on Samsung Research

TRICKVOS: A Bag of Tricks for Video Object Segmentation

TRICKVOS: A Bag of Tricks for Video Object Segmentation

Learn about our collection of optimization techniques that significantly improve the performance of video object segmentation models.

Read on Samsung Research