Tutorial Selections

Presenting the Tutorial Selection Process for CVPR 2022

Tutorial Chairs , CVPR 2022

We wrote this blog post about the tutorial selection process for CVPR 2022 to provide transparency and reflection. Hopefully, the blog will also help create a consistent yet continuously improving selection process for future tutorials.

We received 47 tutorial proposals in total --- a big thank you to all the fantastic organizers! We have rated the quality, relevance to CVPR, team expertise, and diversity of the proposals following the evaluation criteria written in the Tutorials Call for Proposals. Each proposal received at least two independent ratings from us and external reviewers who mainly took care of the proposals with which we had conflicts of interest (thank you Diane Larlus, Wei-Lun (Harry) Chao for this!).

We then met virtually to discuss, group, and rank the proposals, and accepted 25 tutorials. Decisions were made by taking into account both individual reviews and a few global criteria to achieve a good balance of topics while taking diversity into account. After that, we invited four tutorials to provide comprehensive coverage of the exciting research happening in the CVPR community. Note that this year, we decided to label a particular category of tutorials related to either platforms or industrial products that may have impacts on computer vision research. CVPR has traditionally hosted such tutorials. You may find those below under the label “industry-track”.

Here is the complete list of tutorials for CVPR 2022.

Tutorials	Organizers	full/half day	Type
Affine Correspondences and their Applications in Practice	Dmytro Mishkin, Daniel Barath, Levente Hajder, James Pritts	full	Contributed
A post-Marrian computational overview of how biological (human) vision works	Li Zhaoping	full	Contributed
Sparsity Learning in Neural Networks and Robust Statistical Analysis	Yanwei Fu, Xinwei Sun, Yao Yuan, Wotao Yin	half	Contributed
Imaging Through Atmospheric Turbulence: Theory, Simulation, and Restoration	Stanley Chan, Nicholas Chimitt	half	Contributed
Performance Measures in Visual Detection and Their Optimization	Emre Akbas, Sinan Kalkan, Kemal Oksuz	Virtual Only	Contributed
Beyond Convolutional Neural Networks	Neil Houlsby, Alexey Dosovitskiy, Alexander Kolesnikov, Xiaohua Zhai	half	Contributed
Medical Diagnosis Using Computer Vision: Foundations, Advances, Challenges	Soheila Borhani, Reza Borhani, Aggelos K. Katsaggelos	half	Contributed
Contactless Health Monitoring using Cameras and Wireless Sensors	Wenjin Wang, Xuyu Wang, Shiwen Mao	half	Contributed
Inside Plato’s door: a tour in Multi-view Geometry	Luca Magri, Federica Arrigoni	full	Contributed
Deep Visual Similarity and Metric Learning	Timo Milbich, Jenny Seidenschwarz, Ismail Elezi, Laura Leal-Taixe, Björn Ommer	half	Contributed
Recent Advances in Vision-and-Language Pre-training	Zhe Gan, Linjie Li, Chunyuan Li, Jianwei Yang, Pengchuan Zhang, Lijuan Wang, Zicheng Liu, Jianfeng Gao	full	Contributed
Deep AUC Maximization	Tianbao Yang, Yiming Ying, Mingrui Liu, Harikrishna Narasimhan	half	Contributed
Denoising Diffusion-based Generative Modeling: Foundations and Applications	Karsten Kreis, Ruiqi Gao, Arash Vahdat	half	Contributed
Multimodal Machine Learning	Louis-Philippe Morency, Paul Pu Liang, Amir Zadeh	half	Contributed
Neural Fields in Computer Vision	Yiheng Xie, Towaki Takikawa, Shunsuke Saito, Or Litany, James Tompkin, Vincent Sitzmann, Srinath Sridhar	full	Contributed
Evaluating Models Beyond the Textbook: Out-of-distribution and Without Labels	Liang Zheng, Ludwig Schmidt, Aditi Raghunathan, Weijian Deng	half	Contributed
Vision-based Robot Learning	Michael Ryoo, Andy Zeng, Pete Florence, Shuran Song	half	Contributed
Building and Working in Environments for Embodied AI	Angel Chang, Rui Chen, Jiayuan Gu, Yuzhe Qin, Hao Su, Xiaolong Wang, Fanbo Xiang	half	Contributed
High-degree polynomial networks for image generation and recognition	Grigorios Chrysos, Markos Georgopoulos, Razvan Pascanu, Volkan Cevher	half	Contributed
Human-centered AI for Computer Vision	Bolei Zhou; Olga Russakovsky	half	Contributed
Graph Machine Learning for Visual Computing	Guohao Li, Guocheng Qian, Jesus Zarzar, Silvio Giancola, Ali Thabet, Matthias Muller, Federico Tombari, Bernard Ghanem	half	Contributed
How to get quick and performant model for your edge application. From data to application	Paula Ramos-Giraldo, Zhuo Wu, Yury Gorbachev, Raymond Lo	half	Industry-track
Towards always-on egocentric vision research using Meta’s Aria glasses	Zhaoyang Lv, Edward Miller, Hyo Jin Kim, Chris Sweeney, Jing Dong, Jakob Julian Engel, Michael Goesele, Armin Alaghi, Vincent Lee, Julian Straub, Pierre Moulon, Vasileios Balntas, Prince Gupta, Mingfei Yan, Richard Newcombe, Kris Kitani	half	Industry-track
OpenMMLab: A Foundational Platform for Computer Vision Research and Production	Dahua Lin, Chen Change Loy, Ziwei Liu, Kai Chen	half	Industry-track
Creating and Using Synthetic Data for Computer Vision Applications - Rendered.ai	Nathan Kundtz, Chris Andrews, Matt Robinson, Sam Kulkarni, Dan Hedges	half	Industry-track
Computational Imaging	Laura Waller, Katie Bouman, Aviad Levis	half	Invited
Remote Sensing for Agriculture and Food Security	Hannah Kerner, Catherine Nakalembe	half	Invited
Representation Learning and Algorithmic Fairness	Sanmi Koyejo	half	Invited
Ego4D tutorial (joint with workshop)	Kristen Grauman et al.	half	Invited

What makes a good tutorial proposal?

Overall, a good proposal should address a timely and relevant topic for CVPR with the right scope (neither too narrow, nor too general). It should present a detailed coherent program with some pedagogical value, showing a strong coordination. Diversity and inclusion efforts (in all possible senses) are strongly welcome. There is no general rule on the number of speakers (which may be a single one), except that having too many speakers (sometimes who are not organizers) often hurts the coherence of the program. In order to help organizers of future events, we have listed below a few shortcomings that often led to rejection.

Several proposals were about the same topic (e.g., Vision Transformers). For each topic, we had to accept at most one of them to save space for other research areas despite that the remaining ones were of high quality.

Some proposals came from almost the same set of organizers, and their topics were similar to some extent. We did not accept them all for the sake of diversity.

Some high-quality proposals came from the organizers of previously successful tutorials appearing at CVPR, ICCV, and ECCV within the last three years. We accepted those containing significantly new content from their past tutorials and rejected the remaining ones whose content largely overlapped with their past ones (or whose delta was simply not clear enough in the proposal).

Some tutorial proposals were especially challenging to evaluate because they were typically based on many short invited talks on hot topics from influential researchers. Predictably, these events would attract a good audience due to the talks given by high-profile researchers. The main shortcomings we identified were the lack of tutorial material in the program and the lack of coherence between the talks. In other words, such proposals would have had good chances to be accepted as workshops, but the tutorial and pedagogical nature of the event was hard to assess. Besides, these proposals often lacked a coordination plan describing a clear team effort, with a set of organizers different from the set of speakers. Some of these proposals were accepted, but the overall acceptance rate of the workshop-like tutorial proposals was highly reduced for the previous reasons. We would like to provide the following advice for future applicants: (i) avoid having a huge number of short talks given by different speakers, unless the justification is crystal clear from a pedagogical point of view; (ii) make the speakers active organizers of the event with a clear description of the coordination effort conducted for the tutorial; (iii) if the previous criteria cannot be met, consider submitting instead a workshop proposal.

Some proposals were targeted to industrial platforms or products and were also tricky to evaluate. These proposals aim to demonstrate a platform, a software tool, or hardware equipment, providing lectures and interactive demonstrations. These proposals probably have high relevance to computer vision practitioners, and past CVPR conferences had hosted a few of them. A shortcoming we identified in some of these proposals was the lack of discussion on how the event may impact the research in computer vision. This year, we have chosen to label these proposals as “industry-track” tutorials and we accepted four of them.

We hope the above will be helpful.

We thank and admire all for your contributions to the CVPR community!

We look forward to seeing you in CVPR 2022!

Boqing and Julien