I am studying a PhD in Intelligent Cinematography (Y2/3). My research is focused on Dynamic Neural Radiance Fields for cinematographic applications. The project is supervised by Dave Bull and Pui Anantrasirichai and funded by My World.
I completed an MEng in Electronic Engineering with AI (First Class) at the University of Southampton. My dissertation focused on describing philosophical, legal and social decision making frameworks with machine learning algorithms; supervised by Mark Weal.
The first (comprehensive) review of computer vision research in the context of real video content acquisition for entertainment. To establish a structure, we categorise work by General, Virtual, Live and Aerial production, and within each category we discuss various machine learning applications and their links to other forms production. We also provide category-specific comments on future works and discuss the socail responsibilities for conducting ethical research.
Moving towards robust NeRF evaluation using synthetic datasets for point-based architectures. This focuses on point-quality and biases involved with various training and test camera distributions to derive metrics for scene complexity.
We focus on improving the compactness of 4-D NeRF representations while reducing GPU utilization so that we can run large datasets without high performance compute; with minor drop in performance. See webpage for results.
We provide an overview of Dynamic NeRF and Gaussian Splatting research in the context of cinematography and explore the use of these technologies (Nerfacto, 4D-GS and SC-GS) to produce (very) short film. Topics discussed: (1) Dynamic representations, (2) Articulated models vs Scene-based modelling, (3) Data collection
We produced three unsupervised general trend detection models models which. My part focused on detection using variance-related errors. This allowed for comprehensive evaluation of monotonic and heteroscedastic trends and was capable of detecting various behaviours that classical and SoA approaches can not. Additionally, I investigated multi-scale detection for non-uniformly distributed data using a binary change-point detection algorithm to control the window size for evaluation. Documentation here.
By relating classical ML techniques to well defined philosophical, social and ethical frameworks I developed a machine learning algorithm to resolve casual social dilemas. Given a set of possible outcomes, this: (1) selects the appropriate philosophical/social/ethical model based on generally accepted criteria, (2) simulates each model and derive scores for the various resolutions, (3) applies a weighted fusion to make a final choice, and (4) provides reasoning for the choice by refering to the selected models and scores. Documentation here.
The problem: You are blind, in a maze and fires appear randomly arround you each time you take a step. Can we resolve the maze unsupervised? This looks at deep Q-learning approaches to solve the maze and compares results for Deep-Q, Duelling Deep-Q, Rainbow algorithms. Additionally, I investigated combined experience replay and compared the greedy episilon and Boltzmann eplxoration methods. Code here.
I reproduced and extended the proposed Genetic Algorithm; which focuses on replicating the conditions of grouping and dispersal in bacterial micro-colonies to provide insight into cooperative structures. The extension focused on better representing individuals to allow for imprecisions in their understanding of the simulated environment. Paper and code found here.
A reproduction of, Are Neural Nets Modular? Inspecting Functional Modularity through Differentiable Weight Masks (ICLR '21). We reproduce the proposed tool for inspecting functional modularity and tested it on a range of NN architectures (namely CNNs and RNNs).
We wrote a comparative short paper on evaluating various linear and non-linear strategies for multi-objective sensor distribution for general irregular (agricultural) fields; supervised by Jose M. Pena. This primarily uses unsupervised differential evolution.