2020 Computer Vision and Pattern Recognition (CVPR) Conference – Virtual Event Case Study
With ongoing travel restrictions and public gatherings still prohibited by circumstances surrounding COVID-19, the organizers for the Computer Vision and Pattern Recognition (CVPR) Conference — one of the largest events exploring artificial intelligence (AI), machine learning, computer vision, and deep learning — made the decision to convert the in-person event, originally planned to take place in Seattle, WA, to an all-digital conference experience.
Virtual CVPR20 Challenges
The conference date was scheduled for 14-19 June, and the organizers had complex plans for a hybrid meeting, which included a physical aspect as well as a virtual component. With the pandemic situation still ongoing, many important decisions had to be put on hold until the final confirmation was made to change to a digital event. Official Seattle mandates canceled all large in-person events on March 11, leaving a very short window of time to pivot to virtual.
Vendor selection was an important component of rapidly producing a virtual event of this scale. The vendor selected to build the CVPR20 virtual conference infrastructure backed out on May 11.
Due to the compressed timeline, the organizers rushed to prioritize the main conference.
As a result, they felt they were unable to provide the same level of support to tutorials and workshops
The team felt that they did not meet the high standards they set for diversity, inclusion, and accessibility.
After consultation with industry, the expo was canceled requiring creative virtual opportunities for sponsors and patrons
Virtual CVPR20 Successes
The organizers extended immense gratitude to Terry Boult and his team of students, postdocs, and volunteers.
In a very short timeframe, they built a fully scalable infrastructure with worldwide access. The organizers felt that without Terry Boult and his team, the virtual CVPR20 event would not have successfully taken place.
Their effort received substantial support from Alibaba and AWS. The organizers also recognized Shenghua Gao and Jingyi Yu (both from Shanghai Tech) for helping provide permission to support attendance in China.
Virtual content exceeded expectations
With more than 5,000 papers, keynotes, sessions, workshops, and tutorials amounting to 1,497,800 minutes of virtual learning and discussion, the extensive amount of technical content was successfully delivered. Authors and presenters virtually-delivered presentations and engaged in live Q&A with attendees.
This year’s event included 16,955 authors, which is a 20% increase over 2019, and received 5,865 valid submissions, resulting in a 29% increase over 2019
The audience included 7,600 from more all over the world
CVPR audience comprised of:
Purpose of attending CVPR:
57% presented a paper, poster, or talk at the main conference, workshop, or tutorial
33% attended as an active member of the vision community, but not directly involved with presentations
8% are new to Computer Vision, attended to learn more
CVPR consistently debuts the most significant research in the field year after year.
Three papers from this event had more than 10,000 citations each; and one alone had more than 25,000.
Google Scholar ranks CVPR above all other computer science conferences and journals.
Rapidly converting an event to virtual at the scale of CVPR20 with over 24,963 combined hours of virtual learning and discussion, and an audience of 7,600 from all over the world, brought new challenges executing the content as well as managing the customer experience. Despite the complex venue and time frame, the technical aspects of the event were conducted successfully and the organizers have gained vast knowledge about the process, platforms, and customer experience requirements of a virtual event. A post-event survey was conducted, which exposed important areas that can be reviewed and improved upon for subsequent events.
Upon completion of the first-ever virtual event:
14% responded that they were very satisfied/extremely satisfied with the event;
69% of attendees were satisfied/somewhat satisfied,
Fewer than 16% responded that they were not at all satisfied.
Regarding the dual session (+12hr) Q&A format of the main conference:
52% of the audience were favorable about the dual session/Q&A format
27% said it allowed them to attend when convenient and still talk with authors
25% said It was a good idea but the implementation was lacking as authors were not there to chat.
In regards to the audience impression of the main conference pre-recorded presentations (5 minutes for orals and 1 minute for posters) at the virtual event:
34% replied that they were just right
49% responded that the oral videos were just right, but the posters were too short
14% felt all videos were too short
In terms of the main’s conference’s use of Zoom for Q&A:
Over 51% of respondents felt that it was just right
26% replied that the idea was okay, but they experienced technology problems
11% felt the video Q&A sessions were too short/crowded and they could not get into the discussions
8% replied that the video Q&A sessions were too long
3% were not able to participate because their company prohibited Zoom
Regarding networking results, online networking was not utilized as extensively as the technical content; over 25% tried it and felt that it was not good; 10% that tried it felt that it was good; and over 64% did not use it.
CVPR is the premier annual computer vision and pattern recognition conference, co-sponsored by the Computer Society and the Computer Vision Foundation. With first-in-class technical content, the main program, tutorials, workshops, a leading-edge expo, and attended by more than 7,500people annually, CVPR creates a one-of-a-kind opportunity for networking, recruiting, inspiration, and motivation. For more information about CVPR, visitcvpr2020.thecvf.com.