Jonas Wulff
Hi! I'm a Research Scientist at Xyla, where I'm working on Natural Language Processing.
From 2018-2021 I was a Postdoctoral Researcher with Antonio Torralba at the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT.
My work concentrated on learning, understanding, and modifying the representations present inside of deep neural networks, with the goal of better applying them to hard tasks like video generation and understanding.
Prior to MIT, I did my PhD with Michael Black at the Max Planck Institute for Intelligent Systems in Tübingen, Germany, where I worked on optical flow and scene structure estimation.
Email: j o n a s @ x y l a . c o m | Google Scholar | Github
News
-
10/2022
I'm honored to receive the Koenderink Prize at this year's ECCV for our work on transforming the Sintel Open Movie into a benchmark and dataset for Optical Flow (and more!), together with Daniel Butler, Garrett Stanley, and Michael Black. For some background, Michael wrote an excellent blog post, revisiting Sintel after 10 years. -
09/2022
Our paper on learning visual representations from procedural shaders got accepted to NeurIPS 2022. -
03/2022
I'm co-organizing two workshops at ECCV 2022. The Robust Vision Challenge evaluates low- and mid-level vision algorithms across a spectrum of different datasets to test robustness and real-life usability. The workshop What is Motion for? will serve as a venue to discuss downstream uses and usefulness, representations, and challenges in motion estimation. -
09/2021
I started a new position as a Research Scientist at Xyla, where I will be working on Natural Language Processing. -
09/2021
Our paper on learning representations from noise got accepted to NeurIPS 2021. -
01/2021
Our paper on compositionality in GANs got accepted to ICLR 2021. -
08/2020
To compare cross-dataset performance across various vision tasks, we organized the second Workshop for Robust Vision in conjunction with ECCV 2020. -
07/2020
The benchmark for single-image depth prediction for Sintel is now online.
Please note that the benchmark is currently in beta stage and subject to change.
Publications
Procedural Image Programs for Representation Learning
Manel Baradad, Chun-Fu (Richard) Chen, Jonas Wulff, Tongzhou Wang, Rogerio Feris, Antonio Torralba, Phillip Isola
NeurIPS 2022 (to appear)
Paper (coming soon) |
Website
Improving Inversion and Generation Diversity in StyleGAN using a Gaussianized Latent Space
Jonas Wulff and Antonio Torralba
arXiv Preprint
Paper
Inverting layers of a large generator
David Bau, Jun-Yan Zhu, Jonas Wulff, William Peebles, Hendrik Strobelt, Bolei Zhou, Antonio Torralba
ICLR Workshop on Debugging Machine Learning Models 2019
Paper
Temporal interpolation as an unsupervised pretraining task for optical flow estimation
Jonas Wulff and Michael J. Black
German Conference for Pattern Recognition (GCPR) 2018
Paper
"What is Optical Flow for?": Workshop Results and Summary
Fatma Güney, Laura Sevilla-Lara, Deqing Sun, Jonas Wulff
ECCV 2018 Workshop: What is Optical Flow for?
Paper
Model-Based Optical Flow: Layers, Learning, and Geometry
Jonas Wulff
PhD Thesis, University of Tuebingen
Online version
Optical flow in mostly rigid scenes
Jonas Wulff, Laura Sevilla-Lara, Michael J. Black
IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2017
Paper |
Supplemental Material |
Website |
Code
Slow flow: Exploiting high-speed cameras for accurate and diverse optical flow reference data
Joel Janai, Fatma Güney, Jonas Wulff, Michael J. Black, Andreas Geiger
IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2017
Paper |
Supplemental Material |
Website |
Code
Modeling blurred video with layers
Jonas Wulff and Michael J. Black
European Conference on Computer Vision (ECCV) 2014
Paper |
Supplemental Material |
Website
A fully-connected layered model of foreground and background flow
Deqing Sun, Jonas Wulff, Erik Sudderth, Hans-Peter Pfister, Michael J. Black
IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2013
Paper |
Supplemental Material
A naturalistic open source movie for optical flow evaluation
Daniel J. Butler, Jonas Wulff, Gerit B. Stanley, Michael J. Black
European Conference on Computer Vision (ECCV) 2012
Paper |
Supplemental Material |
Benchmark Website
Lessons and Insights from creating a synthetic optical flow benchmark
Jonas Wulff, Daniel J. Butler, Gerit B. Stanley, Michael J. Black
ECCV Workshop on Unsolved Problems in Optical Flow and Stereo Estimation
Paper
Publications prior to 2012
Correspondence estimation from non-rigid motion information
Jonas Wulff, Thomas Lotz, Thomas Stehle, Til Aach, and Geoffrey Chase
SPIE Medical Imaging 2011
Paper
Modellbasierte Echtzeit-Bewegungsschätzung in der Fluoreszenzendoskpie
Thomas Stehle, Jonas Wulff, Alexander Behrens, Sebasitan Gross, Til Aach
Bildverarbeitung für die Medizin 2010
Paper
Polyp segmentation in NBI colonoscopy
Sebastian Gross, Manuel Kennel, Thomas Stehle, Jonas Wulff, Jens Tischendorf, Christian Trautwein, Til Aach
Bildverarbeitung für die Medizing 2009
Publisher link
Denoising fluorescence endoscopy - A motion compensated temporal recursive video filter with an optimal minimum mean square error parameterization
Thomas Stehle, Jonas Wulff, Alexander Behrens, Sebastian Gross, Til Aach
ISBI 2009
Paper
Dynamic distortion correction for endoscopy systems with exchangeable optics
Thomas Stehle, Michael Hennes, Sebastian Gross, Alexander Behrens, Jonas Wulff, Til Aach
Bildverarbeitung für die Medizin 2009
Publisher link
Classification of colon polyps in NBI endoscopy using vascularization features
Thomas Stehle, Roland Auer, Sebastian Gross, Alexander Behrens, Jonas Wulff, Til Aach, Ron Winograd, Christian Trautwein, Jens Tischendorf, SPIE Medical Imaging 2009.
Publisher link