Dylan Campbell

Dylan Campbell is an AI researcher at the Australian National University whose work connects 3D computer vision with differentiable optimisation in deep learning. He earned a mechatronic-engineering degree at UNSW and a geometric-vision PhD at ANU, then held research fellow posts at ANU's Australian Centre for Robotic Vision and Oxford's Visual Geometry Group. His research includes a globally optimal camera-pose method, deep declarative networks that place optimization problems inside trainable models, and trajectory attention for video. In 2022, he and his coauthors improved the backward pass for robust vector pooling and optimal transport; in 2024, an ARC DECRA funds his generative-vision project, and as of October 2026 he is a Senior Lecturer and ARC DECRA Fellow at ANU.

Now at: Australian National University

Timeline

2012

Mechatronic engineering at UNSW

He earns a BE in Mechatronic Engineering with honours from the University of New South Wales in 2012.
2017

Globally optimal camera pose

In 2017, he coauthors a method that jointly estimates camera pose and 2D–3D feature correspondences, using branch-and-bound to guarantee a globally optimal solution.
2018

A geometric vision PhD

He completes a PhD at ANU in 2018 on geometric vision problems at Data61/CSIRO and NICTA, under Lars Petersson, Laurent Kneip and Hongdong Li. In August 2018, he joins ANU's Australian Centre for Robotic Vision as a Research Fellow.
2019

Deep Declarative Networks

In 2019, he coauthors a paper that defines network layers as solutions to optimization problems and uses implicit differentiation to backpropagate gradients, enabling end-to-end learning.
2021

Trajectory attention for video

In 2021, he coauthors “Keeping Your Eye on the Ball: Trajectory Attention in Video Transformers,” which introduces a drop-in block that aggregates information along motion paths. He works as a Research Fellow in Oxford's Visual Geometry Group under Andrea Vedaldi and João Henriques.
2022

Efficient implicit gradients

In 2022, he and four coauthors study deep declarative networks for robust vector pooling and optimal transport, using problem structure to make backward-pass computations more efficient in time and memory.
2024

Worlds unseen DECRA funding

In 2024, an ARC Discovery Early Career Researcher Award provides $472,431 for his project “Worlds unseen: Inferring hidden structure with generative vision models” at ANU.
As of October 2026

Senior Lecturer at ANU

As of October 2026, he is a Senior Lecturer and ARC DECRA Fellow in ANU's School of Computing, where he convenes VComp and the PhB program. His work spans computer vision, optimisation, machine learning and robotics, with particular expertise in 3D vision and optimisation for deep learning.