Ravi Garg

Ravi Garg is a computer vision and machine learning researcher whose work introduced unsupervised learning frameworks for single-view depth and egomotion estimation. He earned a PhD in 2014 under Lourdes Agapito, with a dissertation on dense motion capture of deformable surfaces from monocular video, and then spent six years as a Senior Research Associate with Ian Reid at the University of Adelaide. A 2016 ECCV paper with Vijay Kumar BG, Gustavo Carneiro and Ian Reid trained a depth-prediction network without annotated ground-truth depths, and a 2018 CVPR paper learned depth and visual odometry together from stereo sequences. From 2020 to 2023 he was a Senior Applied Scientist at Amazon Research. Since 2023, he is a Future Making Fellow at the Australian Institute of Machine Learning at the University of Adelaide, where his interests are 3D reconstruction, visual reasoning and generative AI for robotics and automation.

Now at: University of Adelaide
Badges: Bridge
Boomerang

Timeline

2009–2014

A PhD at Queen Mary

He studies at Queen Mary University of London from 2009 and completes his PhD in 2014 under Prof. Lourdes Agapito. His dissertation covers dense motion capture of deformable surfaces from monocular video.
2013

Variational non-rigid reconstruction

In 2013, he coauthors a CVPR paper with Anastasios Roussos and Lourdes Agapito on dense 3D reconstruction of non-rigid surfaces from a monocular video sequence. The method estimates a low-rank smooth 3D shape for every frame through trace norm minimization and total-variation regularization, and needs no prior models or shape templates.
2014–2020

Research associate at Adelaide

From 2014 to 2020, he is a Senior Research Associate at the University of Adelaide, where he works for six years with Prof. Ian Reid. His research puts machine learning into real-time localisation and mapping and introduces unsupervised learning frameworks for single-view depth and egomotion estimation.
2016

Unsupervised depth from image pairs

In 2016, he is first author, with Vijay Kumar BG, Gustavo Carneiro and Ian Reid, of an ECCV paper that trains a convolutional network for single-view depth prediction without annotated ground-truth depths. The network learns in the manner of an autoencoder from image pairs with small, known camera motion, such as stereo pairs, and reaches performance comparable to state-of-the-art supervised methods on KITTI after training on less than half of that dataset.
2018

Depth and odometry from stereo

In 2018, he coauthors a CVPR paper with Huangying Zhan, Chamara Saroj Weerasekera, Kejie Li, Harsh Agarwal and Ian Reid. It trains single-view depth and visual odometry together from stereo sequences, adds a deep-feature warping loss, and outperforms earlier learning-based methods on the KITTI driving dataset in both tasks.
2020–2023

Applied scientist at Amazon

From 2020 to 2023, he is a Senior Applied Scientist at Amazon Research, working with Prof. Anton van den Hengel. In 2022, he is among the authors of Retrieval Augmented Classification, a CVPR paper that adds a retrieval module to image classifiers for long-tail recognition.
Since 2023

Future Making Fellow at Adelaide

He becomes a Future Making Fellow at the Australian Institute of Machine Learning at the University of Adelaide in 2023, with interests in 3D reconstruction, visual reasoning and generative AI for robotics and automation. In 2024, he coauthors Invertible Neural Warp for NeRF with Shin-Fang Chng, Hemanth Saratchandran and Simon Lucey, which models camera poses as learnable rigid warp functions and uses an invertible neural network. In December 2025, he coauthors G3Splat, a framework that places differentiable geometric priors on predicted 3D Gaussians for generalizable splatting.