CV

Computer Vision Research Engineer: 3DGS & Diffusion Avatars

Signapse

• Trained mesh-anchored Gaussian Splatting avatars that render signing in real time from any viewpoint
• Worked on a video diffusion pipeline that refines 3DGS renders into photorealistic avatar video
• Fitted a merged SOMA body and FLAME head model to multi-view captures of a full signer dataset
• Designed a portable multi-view capture rig that recorded 10,000 signs on site in 4 months

PhD Researcher

University of Surrey, CVSSP

• Leading multi-view capture and large-scale 3DMM fitting for the lab's sign language datasets
• Multi-node training on NVIDIA and AMD clusters; wrote two funded Isambard-AI proposals (20,000 GPU-hours)
• Contributing to SLTK, the lab's open-source sign language toolkit

PhD

University of Surrey

Computer Vision

• Expected graduation: early 2028
• Thesis (working title): Preserving Face, Body, and Hand Information in Sign Language Production
• Supervised by Prof. Richard Bowden and Dr. Özge Mercanoğlu Sincan

BEng (Hons), First Class

University of Surrey

Electronic Engineering

• Thesis: real-time photorealistic avatar driven from monocular video (MediaPipe, MetaHumans)

M3T: Discrete Multi-Modal Motion Tokens for Sign Language Production

BMVC

Separate discrete tokens for body, hands, and face (SMPL-X + FLAME), generated from text by a transformer; state-of-the-art 3D sign motion that keeps the facial expressions earlier methods lose.

SignSparK: Efficient Multilingual Sign Language Production via Sparse Keyframe Learning

ECCV

Generates full 3D signing from sparse keyframes with flow matching, across four sign languages, rendered with photorealistic Gaussian Splatting avatars.

VisualSpeaker: Visually-Guided 3D Avatar Lip Synthesis

ICCV Workshops, Oral

Trains photorealistic 3D avatar lip motion through differentiable Gaussian Splatting renders and a lip-reading perceptual loss, giving clearer, more readable mouth shapes.