Dylan Campbell
The Australian National University
School of Computing
About Me
I am a lecturer (US/Europe equivalent: Assistant Professor) in the School of Computing at the Australian National University. Previously, I was a Research Fellow of the Visual Geometry Group at the University of Oxford, where I was supervised by Andrea Vedaldi and João Henriques. Before that, I was a Research Fellow of the Australian Centre for Robotic Vision and ANU, where I was supervised by Stephen Gould. I received my PhD from ANU, supported by Data61/CSIRO, working on geometric vision problems under the supervision of Lars Petersson, Laurent Kneip and Hongdong Li. I hold a BE in Mechatronic Engineering (Hons) from the University of New South Wales, with an honours thesis in robotics completed under the supervision of Mark Whitty.
Research
I have broad research interests within computer vision, optimisation, machine learning, and robotics, with particular expertise in 3D vision and optimisation for deep learning. I have investigated problems of geometric sensor data alignment (including camera localisation, simultaneous localisation and mapping, structure from motion, and optical flow), 3D representations (including neural radiance fields), and differentiable optimisation layers (inserting constrained optimisation problems into deep learning systems). Current topics of interest include discovering and exploiting symmetries in data to share information across long-range physically-motivated connections, and optimisation in deep learning for training neural networks efficiently with respect to time and the quantity of data.
Information for prospective research students
I am always looking for highly motivated students who are interested in conducting research with me. To see what type of work I do, you can read some of my selected papers. I encourage students to contact me but please read the following before doing so:
Existing ANU undergraduate and masters students: From all potential applicants, I require an academic transcript, a CV, a sample of research writing, and a proposed research topic. I have numerous projects ideas in computer vision and machine learning, often advertised here, but I am equally happy for you to propose your own project so long as it aligns with my research interests. If I supervise your project, I will expect you to work consistently throughout, not just before meetings or milestones, and to attend regular meetings. Note that there is no expectation that an Honours project will result in a conference or journal publication. Further resources are available here.
Existing ANU PhD students: Email me to set up a meeting and include information about you and your current project.
Potential PhD students: From all potential applicants, I require an academic transcript, a CV, a sample of research writing, and a proposed research topic. You must contact me at least three months before the admissions deadline, and have read the College's PhD admission website that includes the scholarship deadlines. All applications should come through the ANU applications system. You must have good mathematical and computational skills, preferably with prior experience in machine learning and computer vision. You should have a high standard of oral and written English proficiency, strong programming skills in Python, and familiarity with revision control.
Summer and visiting scholars: Generally, I do not supervise visiting students during the northern hemisphere summer (June to August), but I do participate in the Summer Research Scholar program (December to January). Potential visiting PhD students should send me a description of your work, a list of publications or Google Scholar link, and how your trip will be funded.
Research project proposals: As a minimum, a research proposal consists of (a) the research question, (b) how it relates to existing work, (c) a list of datasets for training/testing and any specialist software required, and (d) a timeline with milestones attached to specific dates. The usual approach is to extend some recent work in a small way (~3 months) and then move onto something more ambitious. Proposal writing resources are available here.
Diversity: I welcome and encourage applications from individuals with diverse backgrounds that may be poorly-represented in computing, including those who identify as women, Aboriginal or Torres Strait Islander, or LGBTQIA+; and individuals from regional or rural areas, culturally diverse backgrounds, low socioeconomic backgrounds, and those living with a disability. I'd particularly like to highlight the Elevate scholarship, which aims to foster the inclusion of women in STEM.
News
Why have Stable Diffusion when you can have Stale Diffusion?
Our paper "IMPUS: Image Morphing with Perceptually-Uniform Sampling Using Diffusion Models" has been accepted for the 2024 International Conference on Learning Representations (ICLR). Congratulations to Zhaoyuan, Zhengyang, and co-authors.
Our paper "Rethinking polyp segmentation from an out-of-distribution perspective" has been published in the Machine Intelligence Research journal.
2023:
Honoured to receive the Australian Pattern Recognition Society (APRS) Early Career Researcher Award 2023 at DICTA 2023 and for being invited to deliver an award talk "Exploring Diffusion Data Spaces".
Our paper "Ray Deformation Networks for Novel View Synthesis of Refractive Objects" has been accepted for the 2024 Winter Conference on Applications of Computer Vision (WACV). Congratulations to Weijian and co-authors.
Our paper "SCENES: Subpixel Correspondence Estimation with Epipolar Supervision" has been accepted for the 2024 International Conference on 3D Vision (3DV) as an oral presentation. Congratulations to Dominik and João.
Spent three days in Oxford catching up with colleagues and friends, discussing research, and delivering an invited talk - thank you for hosting me!
Delighted to present our posters at ICCV 2023 - thank you to everyone who asked such interesting questions!
Our paper "LoCUS: Learning Multiscale 3D-consistent Features from Posed Images" has been accepted for the 2023 International Conference on Computer Vision (ICCV). Congratulations to Dominik and João.
Our paper "Exploring Predicate Visual Context in Detecting Human–Object Interactions" has been accepted for the 2023 International Conference on Computer Vision (ICCV). Congratulations to Fred and co-authors.
Our review article "Robotic vision for human-robot interaction and collaboration: A survey and systematic review" has been published at ACM Transactions on Human-Robot Interaction. Congratulations to Nicole and co-authors.
Started as a tenure-track lecturer at the Australian National University!
2022:
I'm honoured to be an Outstanding Reviewer recipient at ECCV 2022.
Sadly unable to present our work "SNeS: Learning Probably Symmetric Neural Surfaces from Incomplete Data" at ECCV 2022 due to flight cancellations, but thank you to everyone who asked my co-author Eldar such great questions.
Awarded the Best Presentation Award at the 2022 Rank Prize Symposium on Neural Rendering.
I'm excited to be an invited speaker at the 2022 Rank Prize Symposium on Neural Rendering in Computer Vision!
For more on SNeS, check out my blog post!
Our neural fields paper "SNeS: Learning Probably Symmetric Neural Surfaces from Incomplete Data" has been accepted for the 2022 European Conference on Computer Vision (ECCV) [website] [paper] [code]
It's said that "A 23 MW data centre is all you need" [video]
Our paper "Efficient Two-Stage Detection of Human-Object Interactions with a Novel Unary-Pairwise Transformer" has been accepted for the 2022 Conference on Computer Vision and Pattern Recognition (CVPR) - congratulations Fred!
Our article "Geometry-guided street-view panorama synthesis from satellite imagery" has been published at IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI). Congratulations Yujiao and co-authors.
Appeared on the Talking Papers podcast to discuss deep declarative networks - thanks for having me Itzik!
2021:
Our paper "Exploiting Problem Structure in Deep Declarative Networks: Two Case Studies" has been accepted for the 2021 AAAI Workshop on Optimal Transport and Structured Data Modeling - check it out for a nice application of DDNs to optimal transport
Presented "Keeping Your Eye on the Ball: Trajectory Attention in Video Transformers" at NeurIPS 2021 (virtually...) - thank you to everyone who asked such interesting questions at the oral and poster sessions
Our paper "Learning to estimate hidden motions with global motion aggregation" has been accepted for the 2021 International Conference on Computer Vision (ICCV) - congratulations Zac!
Our paper "Spatially Conditioned Graphs for Detecting Human–Object Interactions" has been accepted for the 2021 International Conference on Computer Vision (ICCV) - congratulations Fred!
Our paper "Keeping Your Eye on the Ball: Trajectory Attention in Video Transformers" has been accepted for the 2021 Conference on Neural Information Processing Systems (NeurIPS) as an oral presentation (0.6% acceptance rate)
Our article "Deep declarative networks" published at IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)
2020:
Joined the Visual Geometry Group at the University of Oxford as a Research Fellow under the supervision of Andrea Vedaldi and João Henriques.
Our Deep Declarative Networks tutorial at ECCV2020 has concluded, but leaves behind an excellent set of reference videos for anyone interested in learning about the topic. Thank you to Itzik for taking the lead on organising it and to all the speakers for their fantastic contributions!
Our paper "Solving the Blind Perspective-n-Point Problem End-To-End With Robust Differentiable Geometric Optimization" has been accepted for the 2020 European Conference on Computer Vision (ECCV) as an oral presentation (2% acceptance rate)
Our paper "Deep View Synthesis From Colored 3D Point Clouds" has been accepted for the 2020 European Conference on Computer Vision (ECCV) - congratulations Wayne and Zhenbo!
The Deep Declarative Networks workshop at CVPR2020 was a great success, with over 100 attendees contributing to the discussion of this emerging field of research - thanks to all who helped organise the event and to our fantastic speakers! Watch the talks here.
Our paper "Inferring Temporal Compositions of Actions Using Probabilistic Automata" has been accepted for the 2020 Conference on Computer Vision and Pattern Recognition (CVPR) workshop on Compositionality in Computer Vision (CICV 2020) - congratulations Rodrigo!
Our paper "Where am I looking at? Joint Location and Orientation Estimation by Cross-View Matching" has been accepted for the 2020 Conference on Computer Vision and Pattern Recognition (CVPR) - congratulations Yujiao!
Deep Declarative Networks Workshop (CVPR2020)
The Deep Declarative Networks workshop was held at CVPR2020 and consisted of 6 invited talks and 5 accepted papers with oral presentations. Watch the talks here. Read the papers here.
2019:
We're running a tutorial on Deep Declarative Networks at ECCV 2020 - details to come
We're running a workshop on Deep Declarative Networks at CVPR 2020 - details to come
Reference code and tutorials for deep declarative networks are now available on GitHub
Our paper "The Alignment of the Spheres: Globally-Optimal Spherical Mixture Alignment for Camera Pose Estimation" has been accepted for the 2019 Conference on Computer Vision and Pattern Recognition (CVPR)
2018:
Won 1st prize in the IEEE Australia Council Postgraduate Student Paper Competition for our TPAMI article “Globally-Optimal Inlier Set Maximisation for Camera Pose and Correspondence Estimation”
Accepted a position as Research Fellow with the Australian Centre for Robotic Vision and the Australian National University under the supervision of Stephen Gould
Our article "Globally-Optimal Inlier Set Maximisation for Camera Pose and Correspondence Estimation" has been published in Transactions on Pattern Analysis and Machine Intelligence (TPAMI)
Commenced research project into heart rate detection using semi-supervised deep learning
Completed and submitted PhD thesis "Robust and Optimal Methods for Geometric Sensor Data Alignment"!
Submitted article on camera pose estimation to Transactions on Pattern Analysis and Machine Intelligence (TPAMI)
2017:
Our paper "Globally-Optimal Inlier Set..." etc has won a Marr Prize Honourable Mention at ICCV 2017!
Source code for GOPAC released online
Moved site to Black Mountain Laboratories! The reality is more prosaic than the name unfortunately...
Our paper "Globally-Optimal Inlier Set Maximisation for Simultaneous Camera Pose and Feature Correspondence" has been accepted for the 2017 International Conference on Computer Vision (ICCV) as an oral presentation (2% acceptance rate)
2016:
Source code for GOGMA released online
Our paper "GOGMA: Globally-Optimal Gaussian Mixture Alignment" has been accepted for the 2016 Conference on Computer Vision and Pattern Recognition (CVPR) as a spotlight presentation
Our article "Go-ICP: A Globally Optimal Solution to 3D ICP Point-Set Registration" has been published in Transactions on Pattern Analysis and Machine Intelligence (TPAMI)
2015:
Source code for Support Vector Registration (SVR) released online
Our paper "An Adaptive Data Representation for Robust Point-Set Registration and Merging " has been accepted for the 2015 International Conference on Computer Vision (ICCV) [spotlight presentation]