View on GitHub

Personal Page

Dominic Jack

Computer Science PhD student, Queensland University of Technology

I’m a final year PhD student and lecturer at QUT in Brisbane, Australia. My research is focused on deep learning methods for computer vision, for which I mostly use Tensorflow and Python. I’m currently looking at a variety of 3D inference tasks and representations.

When I’m not cleaning data researching or teaching, I enjoy playing with data visualization tools for the web using dart. I’m currently working on a computer algebra system in dart to help break down complex mathematical processes - think a simpler wolfram alpha’s show steps button, but free and running client-side.


IGE-Net: Inverse Graphics Energy Networks for Human Pose Estimation and Single-View Reconstruction (CVPR 2019)

Paper | Code | Poster

Inferring 3D scene information from 2D observations is an open problem in computer vision. We propose using a deep-learning based energy minimization framework to learn a consistency measure between 2D observations and a proposed world model, and demonstrate that this framework can be trained end-to-end to produce consistent and realistic inferences. We evaluate the framework on human pose estimation and voxel-based object reconstruction benchmarks and show competitive results can be achieved with relatively shallow networks with drastically fewer learned parameters and floating point operations than conventional deep-learning approaches.

Learning Free-Form Deformations for 3D Object Reconstruction (ACCV 2018)

Paper | Code

We train a standard convolutional network to learning free form deformation parameters to reconstruct 3D meshes from single images. The network simultaneously learns to deform multiple known templates and choose an appropriate template for the query image.

Adversarially Parameterized Optimization for 3D Human Pose Estimation (3DV 2017)

Paper | Code

We propose inferring 3D pose from monocular images by searching over the latent feature space of a GAN generator to find feasible 3D poses that match 2D observations. Results indicate that tiny networks can achieve competitive results.

Tensorflow Devpost Competition Entry

My work on point clouds recently made the front page of winning entries to tensorflows Devpost competition (24th out of 675, bottom right). Read a brief summary here, check out the slides, read a brief technical report or dive into the code (warning: this is still under active development - things will be rocky).

Tensorfow Repositories

Reproduced Work

Dataset Repositories

Data IO and cleaning is a necessary evil of almost all machine learning research. I maintain the following repositories for loading/manipulating publicly available datasets.


As well as research, I’m also involved in teaching maths at QUT. I’ve recently done some experimentation with reveal.js but since I much prefer dart over javascript I’ve wrapped a number of maths and data visualization related javascript libraries.