Advice for First-Time Computer Vision Conference Attendees

Email  /  CV  /  Github  /  LinkedIn  /  Twitter

And that's a wrap! I just finished my attendance at my first ever academic conference! And not only was it my first conference, it was also my first virtual conference. At first glance, CVPR2021 was a formidable experience due to two factors: the virtual format and the sheer number of attendees and papers. With more than 7000 attendees, it can be uncomfortable to be present at an event with so many people. Furthermore, the online dimension of the conference made planning an even more daunting task. With so many papers and workshops to choose from, it remained hard to decide which events were actually worth the time to visit.

In this post, I'll explain the things that I enjoyed about this conference, the things that I wish I would have done differently, and ways for you to plan for your next conference attendance 😀

A virtual conference, more so than a physical one, allows for the ultimate freedom to choose which events to attend. With more than 3000 accepted papers spread across 11 paper sessions, it may seem daunting to choose what you want to see.

Plan to attend the conferences in your specific field of research

Instead, just create a curate list of papers that are in your problem domain first. While you may not be able to cover all of them, or if some of them occur at a session time in the middle of the night for you, this will help narrow down what you actually want to see. Watch the pre-recorded videos ahead of time. When you visit the actual session, ask good questions for the authors. I found that for quite a few authors that I met, it was also their first experience with CVPR. For example, I curated a list of open-set recognition (OSR) and open world theory papers that I wanted to see, and I met some new researchers that I was completely unaware of that had a strong understanding of the theory and were able to help guide my understanding of the subject matter.

Find interesting and meaningful topics that still somewhat align with your research topics

Once you have a list of the papers that you want to attend, consider adding the topics that intrigue and/or interest you. For example, I really wanted to learn more about autonomous vehicles while at this conference, so I attended one (of the four) workshops on autonomous vehicles and scanned for papers that seemed to somewhat align with my aforementioned research area.

Try to visit at least one award-winning paper

For example, I visited the paper session GIRAFFE: Representing Scenes As Compositional Generative Neural Feature Fields to see some of the improvements in the state-of-the-art. While this paper is not really in my direct field of expertise, it's valuable to learn from top researchers to see how they conduct their work.

I wanted to write a quick word about workshops too, because they seem to get overlooked because of the prestige of the main conference proceedings, alongside a garden variety of other factors. In my opinion, they remain one of the best places to connect with other experts on super niche concepts, often with highly specific applications. Just for brevity, the list of CVPR 2021 workshops is long and full of interesting topics. Better yet, they don't conflict with any of the main paper sessions. While many of them overlap with one another, I think it's a good idea to attend some - especially if you are bored on a weekend afternoon! 😝

From the perspective of someone who had a workshop paper this year, giving the poster session was a valuable experience. I was able to meet people who were interested in my work and learned to better articulate my research to an audience in a live, synchronous fashion. The poster session allows for continuous involvement and feedback for the audience, which remains valuable from a communications perspective.

This remains the hardest aspect of the normal conference to capture in a virtual setting. Platforms used like GatherTown and Gatherly attempt to simulate the experience of walking up to someone at a cocktail hour, but the effect remains muted without human cues beforehand like smiling, eye contact, and microexpressions.

As a student, I definitely recommend attending the student-curated networking sessions. These have guest speakers who answer questions on a variety of topics. However, I'd caution you to not spend too much time listening to just one speaker. While it may seem great to connect with a speaker, there are often 16 (and never more than 16!!) other students also there, so it remains really quite hard to form a personal connection that you could potentially obtain in an in-person setting. Instead, I'd recommend meeting other students at the networking session itself. From that perspective, I was able to at least connect on LinkedIn with a few of them, and I think that learning how other students from other universities and other parts of the world remain a more valuable experience.

Again, that's just my advice from my personal experience with the networking session. You might receive more value at the actual talks during them, but I think that it's important to meet other students.

A quick word on sponsors

The sponsor sessions remained more of just a pre-recorded advertising session (although the gift packages seemed nice). The real chance to meet recruiters was at the exhibitor section. From there, you could meet them at designated scheduled times. While some companies provided rather fluffy presentations and panel sessions, many also provided special CVPR2021 links for applications for internships and full-time roles.

Overall, I enjoyed this event, but I also love to see exactly what it would look like in-person. According to the TC-PAMI meeting at the event, future conferences might follow a hybrid model in which attendees can view more archivable content while still having a choice to see a conference in person. Another option remains to segment a big conference into smaller sub-sections at different time points for in-person activities.

Regardless of the future, I hope that this guide remains helpful to you and your next conference experience.

source code