James Gardner

University of York

I am a third-year PhD student at The University of York, specialising in computer vision and machine learning for 3D scene understanding. Supervised by Dr William Smith, my research focuses on neural-based vision and language priors in inverse rendering and scene representation learning. I'm particularly interested in neural fields, generative models, 3D computer vision, differentiable rendering, geometric deep learning, multi-modal models, and 3D scene understanding in general. My research has been recognised with publications at prestigious conferences including NeurIPS and ECCV.  

Currently, I am working as a research fellow on the ALL.VP project, funded by BridgeAI and Dock10, developing relightable green screen performance capture using deep learning and inverse rendering techniques. This work aims to provide greater creative control to film and TV productions without requiring expensive LED volumes or post-production.  

I hold an MEng in Electronic Engineering from The University of York, for which I was awarded the IET Prize for outstanding performance and the Malden Owen Award for the best-graduating student on an MEng programme.

A description of James' research:

My research lies at the intersection of computer vision, machine learning, and 3D scene understanding, with a particular focus on neural-based approaches and the integration of vision and language priors. My work spans a range of topics including neural fields, generative models, differentiable rendering, and geometric deep learning. A key theme in my research is the use of 3D inductive biases for inverse rendering, addressing challenges such as illumination estimation, albedo/geometry disentanglement, and shadow handling in complex outdoor scenes. I've made contributions in creating a rotation-equivariant neural illumination model and spherical neural models for sky visibility estimation in outdoor inverse rendering. 

Additionally, my work extends to learning rotation-equivariant latent representations of the world from 360-degree videos, aimed at advancing the field of 3D scene understanding and developing models with an understanding of core physical principles such as object permanence. Through my research, I aim to build computer systems capable of deeply comprehending the 3D world, utilising self-supervised, generative, and non-generative approaches to push the boundaries of what's possible in computer vision and scene representation learning.