Adrian

Azzarelli

Neural 3-D reconstruction, Gaussian Splatting and real-time rendering for filmmaking and immersive experiences.

About

Photo of Adrian Azzarelli

I do R&D on neural 3-D reconstruction for filmmaking applications. I've worked on dynamic NeRFs and Gaussian Splatting for Metaverse (6DoF) entertainment, relightable Splatting for Virtual Production settings and now I work on underwater dynamic, large-scale and relightable Splatting for immersive underwater experiences. My work involves designing and directing capture workflows (working intimately with film crews), R&D on end-to-end and scalable machine learning pipelines, and R&D on interactive rendering pipelines for scenes containing splats, meshes, colliders, etc.

I studied my PhD on Applications of 3-D Reconstruction in Filmmaking settings, at the University of Bristol; and my MEng on Electronic Engineering with AI at the University of Southampton.

If you are curious about my work, want to suggest a project/collaboration or looking to hire, please email me: aazzarellib@gmail.com

To download my CV click here!

Current Work

Technical Specialist @ BLUEAI

Contracted May-Oct 2025/26 | Visual Information Lab, Universtiy of Bristol

BLUEAI brings together photo-real underwater reefs (Gaussian Splatting) and game-engine design (CG fish mechanics), to exhibit a compelling Virtual Reality experience for reef preservation. The product: Users are placed on a moving chair, strapped to an Apple Vision viewing the live reef at 4K per eye at 90FPS, with a joy stick to "swim" around with full 6DoF.

I am responsible for everything from capture to game engine design and implementation. This covers: datacollection/caputre workflows, pose estimation & prior-generation workflows, large-scale, underwater 3-D reconstruction pipelines (4D Splatting), VFX/splatting-shaders, rendering optimization, CG fish mechanics, and more...

Research

2025/6

Relightable Gaussian Splatting for Virtual Production Using Image-Based Illumination

SIGGRAPH ASIA 2026

2025/6

Splatography: Sparse multi-view dynamic Gaussian Splatting for film-making challenge technical

International Conference on 3D Vision 2026 (3DV)

2024/5

ViVo: A Dataset for Volumetric Video Reconstruction and Compression dataset

Special Issue on Volumetric Video Reconstruction and Compression, T-CSVT, IEEE 2026

2024

Exploring Dynamic Novel View Synthesis Technologies for Cinematography review & application

arXiv

2023

WavePlanes: A compact Wavelet representation for Dynamic Neural Radiance Fields technical

arXiv

2022/5

Reviewing Intelligent Cinematography: AI research for camera-based video production review

Springer Nature: Artificial Intelligence Review, Vol 58 Issue 4 (+40 pages)

Fun Side Projects

Rendering Gaussian Splats as Orthonormal Voxels

Python Code | Raster/CUDA Code

The idea: Can we take a generic .PLY splat and render out the scene as a set of othonormal voxels. This is an appearance effect, implemented in CUDA/C++ as a wrapper to Python. This bridges two algorithms: AABB-splat sampling and SLAB-based rendering for splat-like voxels (i.e. voxels with opacity and view-dep color).

This project was mostly intended to improve my CUDA/C++ knowledge and skills. Hence, it was implemented with restrictions on using LLMs for coding assistance.

Surface Traversal on Gaussian Splatting Models for Clothing Design (Pattern Making)

Python Code

The idea: Can we develop an algorithm to trace the straightest path between two points on the surface of a 3D-GS asset? If you know splatting, the idea of a "surface" is ambiguous, so we define it as a minimum density value. The code implements picking and a procedural ray-marching algorithm that uses the shadow of the line A-B projected radially outwards to carve the straightest line across the surface.

Why the "straightest" line? Becasue, I am building a clothing-pattern designer app that uses G-Splatting to produce metric and appearance accurate models. In pattern making, straight lines are essential for accurate tailoring. Hence, running a different optimizer (e.g. smallest distance or fastest render) would not work for pattern making.

This project was mostly intended to improve my knowledge of graphics concepts applied to new 3-D representations.

Surface Traversal on Tri-Meshes (Grease Pencil)

C#/OpenGL Code

The idea: This was essentially a pre-cursor to the Splatting project above. Same objective but more functionality. The underlying goal way to simulate the infamous Grease-pencil, which is actively used in high-end animated films.

This project was mostly intended to improve my knowledge of graphics concepts applied to well known 3-D representations.

Education

Electronic Engineering With AI, MEng

Sep-Sep 2018/22 | Universtiy of Southampton | First Class
Thesis supervised by Mark Weal

Languages & Libraries

Spoken Languages

EnglishFrenchItalian

I am French and Italian, and grew up in London, UK. I can speak English and French with high profficiency. My Italian is casual.

Programming

Currently using

PythonC++H/GSLSJavaScript

Prior experience with

C# (with Unity)embedded-CHSLSJuliaMatLab

My MEng required me to learn objective-C, C++, Python and MatLab applied to various domains including embedded systems, image processing and robotics. My PhD required further developed my Python skills. I have also engaged in some personal projects (e.g. following CatLike Tutorials) to develop my understanding of C#, HSLS and C++ for graphics applications. I have also tinkered with Julia, JS and other languages for some personal tools.

Software & Libraries

BlenderUnityUnreal EngineProcreate ClipStudioPaintPhotoshopDaVinci Resolve CUDA/Optix (RTX code)PyTorchNumpy/SciPy/CuPyPython-C++ wrappers

I have experience using Blender, Unity, Unreal, Procreate CSP, Photoshop and DaVinci for personal artistic projects. Whether its for a cool effect or smooth animation, I love exploring new ways to develop my artistic abilities. I also have work/coding experience with Blender-Python, Unreal-C++ and Unity-C#.