Ronald Clark

I am a research fellow at Imperial College London where I work on 3D vision and machine learning and was recently fortunate enough to receive an Imperial College Early Career Fellowship. I did my PhD at the University of Oxford Department of Computer Science where I was funded by an EPSRC scholarship. Before coming to the UK I did my BSc and MSc in electrical engineering at the University of the Witwatersrand in South Africa.

 /   /   

Research

My research centers around 3D machine perception which is needed to enable mobile devices to model, explore and understand their surroundings. I am particularly interested in ways in which deep learning can be used alongside traditional mathematical and geometrical models to unlock a new level of performance in spatial machine perception. My current work focusses on how recurrent, convolutional neural networks can be used to create consistent, dense, semantically annotated reconstructions of the world. I also have a keen interest in computer graphics and animation.

News
  • Sep, 2019: 1 paper accepted at NeurIPS'19 as a spotlight!
  • Jul, 2019: 2 papers accepted at ICCV'19.
  • Jul, 2019: Invited Talk at the BMVA technical meeting on Geometry and Deep Learning.
  • March, 2019: Our 2nd Workshop on Deep Learning for Visual SLAM will be held at ICCV'19!
  • Feb, 2019: Invited Talk at MeetAI London.
  • Jan, 2019: Honoured to be an early-career ICRF recipient.
  • Dec, 2018: Invited Talk at Microsoft Research Cambridge.
  • Dec, 2018: Co-organizing the CVPR'19 Workshop on Deep Learning for Visual Semantic Navigation
  • Jun, 2018: CVPR Best Paper Honourable Mention Award
Publications


Learning Object Bounding Boxes for 3D Instance Segmentation on Point Clouds
Bo Yang, Jianan Wang, Ronald Clark, Qingyong Hu, Sen Wang, Andrew Markham, Niki Trigoni
Neural Information Processing Systems (NeurIPS), 2019 (Spotlight)
arxiv / bibtex

A simple and accurate method for 3D instance segmentation on point clouds.

X-Section: Cross-section Prediction for Enhanced RGBD Fusion
Andrea Nicastro, Ronald Clark, Stefan Leutenegger
IEEE International Conference on Computer Vision (ICCV), 2019 (Oral)
arxiv / bibtex

We propose a model to predict the thickness of objects and use it to improve the accuracy of multi-view RGB-D reconstruction.

Balancing Reconstruction Quality and Regularisation in ELBO for VAEs
Shuyu Lin, Stephen Roberts, Niki Trigoni, Ronald Clark
arxiv preprint
arxiv

A principled approach for balancing the reconstruction-regularisation tradeoff in VAE models.

Learning Meshes for Dense Visual SLAM
Michael Bloesch, Tristan Laidlow, Ronald Clark, Stefan Leutenegger, Andrew J. Davison
IEEE International Conference on Computer Vision (ICCV), 2019
arxiv / bibtex

Meshes for multiview reconstruction.

Learning Semantically Meaningful Embeddings Using Linear Constraints
Shuyu Lin, Bo Yang, Robert Birke, Ronald Clark
International Conference on Computer Vision and Pattern Recognition (CVPR-W) Workshops, 2019
arxiv / bibtex

We investigate the use of invertible decoders for learning a semantically meaningful latent space.

WiSE-ALE: Wide Sample Estimator for Approximate Latent Embedding
Shuyu Lin, Ronald Clark, Robert Birke, Niki Trigoni, Stephen Roberts
International Conference on Learning Representations (ICLR) Workshops, 2019
arxiv / bibtex

The reconstruction quality of VAEs can be significantly improved by using aggregate posteriors.

Fusion++: Volumetric Object-Level SLAM
John McCormac*, Ronald Clark*, Michael Bloesch, Stefan Leutenegger, Andrew J. Davison
International Conference on 3D Vision (3DV'18), 2018
bibtex

By combining visual SLAM and deep learning we can create object-level maps of the world.

Learning to Solve Non-Linear Least-Squares for Monocular Stereo
Ronald Clark, Michael Bloesch, Jan Czarnowski, Stefan Leutenegger, Andrew J. Davison
European Conference on Computer Vision (ECCV'18), 2018
project page / video / bibtex

We can increase the convergence rate of least-squares optimisers by learning the update steps.

InteriorNet: Mega-scale Multi-sensor Photo-realistic Indoor Scenes Dataset
Wenbin Li, Sajad Saeedi, John McCormac, Ronald Clark, Dimos Tzoumanikas, Qing Ye, Yuzhong Huang, Rui Tang, Stefan Leutenegger
British Machine Vision Conference (BMVC), 2018
project page / video / bibtex

A large rendered dataset of indoor scenes with ground truth semantics, instances, poses and more.

CodeSLAM - Learning a Compact, Optimisable Representation for Dense Visual SLAM   (Best Paper Honorable Mention)
Michael Bloesch, Jan Czarnowski, Ronald Clark, Stefan Leutenegger, Andrew J. Davison
International Conference on Computer Vision (CVPR), 2018
project page/ video / bibtex

Scenes can be represented by the latent space of an autoencoder and optimized for multi-view consistency.

Meta Learning for Instance-Level Data Association
Ronald Clark, John McCormac, Stefan Leutenegger, Andrew J. Davison
Neural Information Processing Systems (NIPS'17) Workshop on MetaLearning, 2018
bibtex

Learning the finetuning process can improve the performance of video object segmentation.

Efficient Indoor Positioning with Visual Experiences via Lifelong Learning
Hongkai Wen, Ronald Clark, Sen Wang, Xiaoxuan Lu, Bowen Du, Wen Hu, Niki Trigoni
IEEE Transactions on Mobile Computing (TMC), 2018
bibtex

Visual localisation can be made more efficient by using predictive, second-order information.

3D Object Reconstruction from a Single Depth View with Adversarial Learning
Bo Yang, Hongkai Wen, Sen Wang, Ronald Clark, Andrew Markham, Niki Trigoni
IEEE International Conference on Computer Vision Workshops (ICCV-W), 2017
code (official) / bibtex

We can use a GAN to learn the conditional distribution of 3D objects given a depth map.

A Deep Spatio-Temporal Model for 6-DoF Video-Clip Relocalization
Ronald Clark, Sen Wang, Hongkai Wen, Andrew Markham, Niki Trigoni
Computer Vision and Pattern Recognition (CVPR), 2017
code (third-party) / bibtex

A model for visual localization that exploits temporal smoothness.

VINet: Visual-Inertial Odometry as a Sequence-to-Sequence Learning Problem
Ronald Clark, Sen Wang, Hongkai Wen, Andrew Markham, Niki Trigoni
AAAI Conference on Artificial Intelligence (AAAI) , 2017
project page / code (third-party) / bibtex

Recurrent Neural Networks can be used to fuse visual and inertial information to estimate the motion of a device.

End-to-End, Sequence-to-Sequence Probabilistic Visual Odometry through Deep Neural Networks
Sen Wang, Ronald Clark, Hongkai Wen, Niki Trigoni
International Journal of Robotics Research (IJRR), 2017
project page / code (third-party) / bibtex

We can do visual odometry using a RNN-CNN model.

DeepVO: Towards End-to-End Visual Odometry with Recurrent Convolutional Networks
Sen Wang, Ronald Clark, Hongkai Wen, Niki Trigoni
bibtex

Conference version of above paper.

Large Scale Indoor Keyframe-based Localization using Geomagnetic Field and Motion Pattern
Sen Wang, Hongkai Wen, Ronald Clark, Andrew Markham, Niki Trigoni
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2016
bibtex

Graph-based SLAM using a magnetometer.

Increasing the Efficiency of 6-DoF Visual Localization Using Multi-Modal Sensory Data
Ronald Clark, Sen Wang, Hongkai Wen, Andrew Markham, Niki Trigoni
IEEE-RAS International Conference on Humanoid Robots (Humanoids), 2016
bibtex

We can improve the efficiency of visual localization by using an informed matching.

Optimization of a Hybrid Energy System through Fast Convex Programming
Ronald Clark, W.A. Cronje, Anton van Wyk
IEEE International Conference on Intelligent Systems Modelling and Simulation (ISMS), 2014
bibtex

Using a convex approximation of the objective function combined with a convex programming solver we can find the optimal configuration of a hybrid energy system in realtime.

System for the recognition of online handwritten mathematical expressions
Ronald Clark, Quik Kung, Anton van Wyk
IEEE International Conference on Computer as a Tool (EUROCON), 2014
bibtex

A system for recognising and solving handwritten equations.

Recognising Handwritten Mathematical expressions using an Ensemble of SVM Classifiers
Ronald Clark, Quik Kung, Anton van Wyk
13th Symposium of the Pattern Recognition Association of South Africa (PRASA/IAPR), 2012
bibtex

Thesis

Visual-inertial odometry, mapping and re-localization through learning.
Ronald Clark
PhD Thesis. University of Oxford.
pdf / bibtex

Academic Service

Reviewer, CVPR, ICCV, BMVC, IJCV, TPAM, ICRA, IROS, RAL

cs188

Teaching Assistant, Computer Animation, University of Oxford (2015,2016,2017)

Assistant Lecturer, Network Fundamentals, Univeristy of the Witwatersrand (2013)


CSS by: jon barron.