|
Towaki Takikawa
I am a first-year Computer Science PhD student at the
University of Toronto, supervised by
Sanja Fidler and
Alec Jacobson.
I also work as a Research Scientist at NVIDIA Research
on the Hyperscale Graphics Research group with
Morgan McGuire.
My research interests focus on computer vision and computer graphics. Specifically, I'm
interested in exploring a suitable representation for 3D geometry that is amenable
to machine-learning driven geometry processing algorithms.
Outside of research, I also have experience working on both
software and hardware for robotics related projects.
I generally enjoy straddling the line between research and engineering.
I come from a small town in Oregon named Corvallis,
but these days I hang out in Canada.
|
|
|
Publications
(* Denotes equal contribution.)
|
|
Neural Geometric Level of Detail: Real-time Rendering with Implicit 3D Shapes
Towaki Takikawa*, Joey Litalien*, Kangxue Yin, Karsten Kreis, Charles Loop,
Derek Nowrouzezahrai, Alec Jacobson, Morgan McGuire, Sanja Fidler
Computer Vision and Pattern Recognition (CVPR), 2021
Oral Presentation
Abstract /
Bibtex /
Project Website
Neural signed distance functions (SDFs) are emerging as an effective representation for 3D shapes. State-of-the-art methods typically encode the SDF with a large, fixed-size neural network to approximate complex shapes with implicit surfaces. Rendering with these large networks is, however, computationally expensive since it requires many forward passes through the network for every pixel, making these representations impractical for real-time graphics. We introduce an efficient neural representation that, for the first time, enables real-time rendering of high-fidelity neural SDFs, while achieving state-of-the-art geometry reconstruction quality. We represent implicit surfaces using an octree-based feature volume which adaptively fits shapes with multiple discrete levels of detail (LODs), and enables continuous LOD with SDF interpolation. We further develop an efficient algorithm to directly render our novel neural SDF representation in real-time by querying only the necessary LODs with sparse octree traversal. We show that our representation is 2-3 orders of magnitude more efficient in terms of rendering speed compared to previous works. Furthermore, it produces state-of-the-art reconstruction quality for complex shapes under both 3D geometric and 2D image-space metrics.
@article{takikawa2021nglod,
title = {Neural Geometric Level of Detail: Real-time Rendering with Implicit {3D} Shapes},
author = {Towaki Takikawa and
Joey Litalien and
Kangxue Yin and
Karsten Kreis and
Charles Loop and
Derek Nowrouzezahrai and
Alec Jacobson and
Morgan McGuire and
Sanja Fidler},
year = {2021},
journal = {arXiv preprint arXiv:2101.10994}
}
|
|
Gated-SCNN: Gated Shape CNNs for Semantic Segmentation
Towaki Takikawa*, David Acuna*, Varun Jampani, Sanja Fidler
International Conference on Computer Vision (ICCV), 2019
Abstract /
Bibtex /
Project Website
Current state-of-the-art methods for image segmentation form a dense image representation where the color, shape and texture information are all processed together inside a deep CNN. This however may not be ideal as they contain very different type of information relevant for recognition. Here, we propose a new two-stream CNN architecture for semantic segmentation that explicitly wires shape information as a separate processing branch, i.e. shape stream, that processes information in parallel to the classical stream. Key to this architecture is a new type of gates that connect the intermediate layers of the two streams. Specifically, we use the higher-level activations in the classical stream to gate the lower-level activations in the shape stream, effectively removing noise and helping the shape stream to only focus on processing the relevant boundary-related information. This enables us to use a very shallow architecture for the shape stream that operates on the image-level resolution. Our experiments show that this leads to a highly effective architecture that produces sharper predictions around object boundaries and significantly boosts performance on thinner and smaller objects. Our method achieves state-of-the-art performance on the Cityscapes benchmark, in terms of both mask (mIoU) and boundary (F-score) quality, improving by 2% and 4% over strong baselines.
@inproceedings{Takikawa2019GatedSCNNGS,
title={Gated-SCNN: Gated Shape CNNs for Semantic Segmentation},
author={Towaki Takikawa and David Acuna and Varun Jampani and Sanja Fidler},
year={2019}
}
|
 |
NVIDIA
Research Scientist
Toronto, ON, Canada
February, 2020 - Present
Mentor: Morgan McGuire
Research on machine learning for computer graphics on the Hyperscale Graphics Research group.
|
 |
WATonomous
Software Director / Mentor
Waterloo, ON, Canada
May, 2017 - May, 2020
Helping the team engineer a fully autonomous Cheverolet Bolt.
Certified by GM as a designated test driver.
Working across the pipeline on perception, prediction, localization, planning.
Managing over 100+ students and 6 different software subteams.
|
 |
Preferred Networks
Research Intern
Tokyo, Japan
May, 2019 - August, 2019
Mentor: Richard Calland, Tommi Kerola
Research on weakly-supervised instance segmentation.
|
|
NVIDIA
Research Intern
Toronto, ON, Canada
January, 2019 - April, 2019
Mentor: David Acuna, Sanja Fidler
Research on representation learning for semantic segmentation.
|
 |
University of Waterloo
Research Assistant
Waterloo, ON, Canada
September, 2018 - May, 2020
Mentor: Yuri Boykov
Research on weakly-supervised semantic segmentation and domain adaptation.
|
|
NVIDIA
Systems Software Engineering Intern
Santa Clara, CA, USA
May, 2018 - August, 2018
Worked in the autonomous vehicles engineering team.
Developed data labeling and data recording systems in C / C++.
Performed hardware verification and validation in the lab.
Helped architect a hardware-based sensor synchronization system.
|
 |
Tulip Interfaces
Software Engineering Intern
Somerville, MA, USA
May, 2017 - August, 2017
Worked on software infrastructure and internal tools.
|
 |
Active911
Software Engineering Intern
Philomath, OR, USA
June, 2016 - August, 2016
Worked on software infrastructure and backend development.
|
 |
Linn Benton Community College
Software Engineering Intern
Corvallis, OR, USA
June, 2015 - August, 2015
Worked on curriculum development for an introductory CS course.
|
 |
Oregon State University
Research Intern
Corvallis, OR, USA
June, 2014 - August, 2014
Mentor: Alex Groce
Research in automated software verification and validation.
|
Custom Gearbox
6 CIM Anodized
Quaternion Blast
CodeDay Portland 2015
|
