Samuel Overington - Curriculum Vitae

Dimension Studio, London UK (Nov 2023 -- Present)

Senior Software Engineer - Machine Learning and Research (Jun 2025) -- Present Software Engineer - Machine Learning and Research

Developing and optimising production-ready machine learning and computer vision pipelines based on state-of-the-art research to power live experiences and virtual production.

Live human matting project for Omega at Paris Olympic Games 2024 [Pytorch, OpenCV, Matting]

Developed and deployed a distributed ML-based video processing pipeline for a live activation (300+ completed renders/day), enabling multi-camera alignment, ML based background matting, and automated editing. Key achievements:

• Engineered a multi-camera temporal synchronisation pipeline utilising global timecodes and Genlock hardware to align four independent video feeds with an LED volume display shutter.
• Architected a distributed, multi-node GPU processing queue across shared network storage, managing state transitions (ready, processing, complete) to scale processing throughput across multiple machines.
• Optimised VRAM utilisation and system throughput by implementing a zero-copy GPU memory strategy, keeping video tensors resident in GPU memory throughout decoding, inference, and encoding stages to meet a strict 5-minute end-to-end processing SLA.
• Developed a deterministic computer vision algorithm to anchor video timelines by localising and detecting a single-frame optical flash event, cross-referencing it with JSON-based finish-line metadata.

Real-time depth & human segmentation in UE [PyTorch, OpenCV, ONNX, TensorRT, Unreal Engine, C++, Segmentation, Depth Estimation, Object Tracking]

Engineered a high-performance Unreal Engine C++ Plugin integrating live video streams with synchronised machine learning models for real-time depth estimation and human segmentation using TensorRT runtime. Key achievements:

• Ported and compiled PyTorch models to ONNX, utilising TensorRT for hardware-accelerated INT8/FP16 inference optimisation within the live execution pipeline.
• Re-architected a model graph to overcome TensorRT compilation constraints on dynamic hidden states, decoupling the encoder-decoder layers and extracting the feature cache aggregation logic from the PyTorch nn.Module to be managed natively in C++.
• Designed a zero-copy GPU memory pipeline, ensuring tensors remained entirely in VRAM between depth inference, segmentation steps, and Unreal Engine texture registration to eliminate CPU-GPU serialisation bottlenecks.
• Optimised high-throughput tensor pre- and post-processing routines to run concurrently, aligning inference latency with the engine’s main render thread to maintain a stable, real-time frame rate.

Semantic segmentation and matting pipeline for virtual production [PyTorch, OpenCV, ML Video Matting]

Leading the development of an automated segmentation pipeline to generate high-quality alpha mattes from raw footage for virtual production. Key achievements:

• Leveraged vision-language models such as BLIP2, GroundingDINO and SAM2 for text-to-object prompting, object detection and segmentation.
• Integrated zero-shot text-to-mask segmentation methods and refined outputs using generative AI models to create precise, production-ready alpha mattes.
• Optimised the output for compositing tools like Nuke and Resolve by creating Cryptomatte files, streamlining the post-production process.

Expedia, London UK (Feb 2022 -- Nov 2023)

Machine Learning Engineer - Growth Marketing & Technology (meta-bidding team)

Architected and scaled data infrastructure for training, deploying, and running high-throughput machine learning prediction pipelines on AWS for meta-bidding and business intelligence.

Arm, Cambridge UK (Nov 2020 -- Jan 2022)

Software Engineer (April 2021) Graduate Engineer

Participated in Arm’s graduate rotation programme, gaining hands-on experience across multiple teams and projects within the machine learning group. Proactively selected rotations to build expertise in machine learning, computer vision, and embedded systems.

• ML Research team, Bayesian DeepLearning for CV (2 months): Led research and implementation of Bayesian models for pixel-level image segmentation, including DeepLabV3 optimisations for Arm hardware.
• Applied ML team, Dynamic hand gesture recognition, and visual wake words (8 months): Lead research engineer conducting a literature review of models and datasets, training and optimising a video understanding model for gesture recognition, and implementing a data generator pipeline using OpenCV for large video datasets. Applying state-of-the-art techniques to enhance performance on low-power, IoT, and embedded Arm IP.
• ML Tooling Team, IPSS-ML (IP Selection Sandbox for ML applications) (4 months): Part of the team developing a middle-ware application to simulate and test ML applications on accelerated (NPU) and regular (M/A class processors) Arm IP using fast- and cycle-models.

Internships:

• Deimos Space UK, Oxfordshire, July – Sept 2019. Computer Vision and Machine Learning research intern
• Yobota, London, Jun – Sep 2018. Software Engineer intern

Selected freelance:

• Stillnessinyoga, Remote, The Netherlands Web developer / Digital content development (June 2016 – 2020).
• LVSC, Web Developer (2013 – 2015)