Home

Contact me

Novel View Synthesis

I specialise in the generation of photorealistic 3D models from an unstructured set of photos or videos. Currently, the two main methods at the forefront of the field are Neural Radiance Fields and Gaussian Splatting.

I've been active in this area of research for many years, and I consistently stay up to date with the latest innovations to cover many possible use-cases of these technologies.

A small selection of my work

Fragment

Fragment is a one-stop-shop to convert videos into photorealistic 3D models with computer vision and machine learning.

Under the hood, the 3D model reconstruction is powered by customised NeRF models, with help of several segmentation networks to clean up unwanted objects.

I built Fragment from the ground up: Frontend with in-browser renderer using React and Three.js, and backend with ML models using Python.

VISIT
Fragment image

All Axis

All Axis creates interactive 3D product photos that are simple to integrate on any website.

I have an ongoing collaboration with All Axis. I built a custom pipeline to convert unstructured photos into a photorealistic 3D model and renders novel images using a custom Gaussian Splatting solution. Additionally, I implemented a suite of internal tools to modify and edit the generated gaussian splats.

After the model is processed, it can be displayed just about anywhere or used to generate additional product photos.

VISIT

TermiNeRF

Ray Termination Prediction for Efficient Neural Rendering

TermiNeRF is a research paper I published as the first author at the 3DV 2021 conference.

Our method decreased the number of neural network evaluations in the original Neural Radiance Fields paper, and sped up the rendering pipeline approximately 14 times.

My Master’s thesis, which served as a basis for the publication, was recognised as a distinguished project by the Department of Computing at Imperial College London.

VISIT

Interested in working together?

Get in touch at hello@martinpiala.com or find me on LinkedIn.

© 2024 Martin Piala. All rights reserved.