ECCV 2022:

In October 2022 I went to ECCV, it’s one of the popular computer vision conference, held every two years. You can read more about the conference itself here. The conference consisted of two days of workshop/tutorials followed by three days of oral/poster presentation, exhibition by companies and couple of keynotes.

• Day 1 was several full/half day workshops organized across several physical and virtual rooms, it was quite difficult to pick one since they were all very interesting (24 page doc program guide for first 2 days: https://eccv2022.ecva.net/files/2022/10/ECCV_2022_WorkshopsTutorials_ProgramGuide_Final_full.pdf)

• I picked “Self Supervised Learning: What is Next?” (SSLWIN) full day workshop discussing future direction of self supervised learning. This is was the most popular workshop/talk in the whole conference with speaker like Alexei Efros and Ishan Misra. You can read more the workshop here.

• Day 2 had both workshop and tutorials with same half or full day schedule. I decided to diversify and attend two half day tutorials

• In the first half I attended tutorial on “Video Synthesis” by Snap Research, the tutorial went over history of this area since GANs in 2014 and work they have done towards generating videos/images in different modalities. In one of the demo they managed to convert a real video into controlled motion (video game like!) and that was really mind blowing to look at. Read more about the tutorial here.

• In the second half I attended tutorial on Neural Volumetric Rendering for Computer Vision OR NERF by google research. NERF seems to be quite effective for photo realistic 3D reconstruction from calibrated images. They also released open source library https://docs.nerf.studio/ so you can actually train, try and look at the results on your own dataset. Abstract for the tutorial can be found here.

• Last 3 days were relatively less hectic with no half/full day commitments, but you can get lost in sea of poster and oral presentation rotating every few hours. If you have time, fell free to scroll through this 56 page doc listing many of these posters and presentations

• There were also a lot of companies present in exhibition like GM, Huawei, Baidu, Samsung, Snap, Rafael, FAANG and other not so known companies. I played a game of chess with robotic arm made by Bosch.

• There were two keynotes in the conference, they invited researchers who are working on stuff relevant to computer vision but are kinda outsider to the field.

• The Science of Ethical Algorithm Design, Michael Kearns: In the keynote, the speaker presented naive algorithm on how you can actually implement ethic in any model. He proposes democratic correction of model in the subset of dataset where model doesn’t behave “ethically”. You can read more about this in the book he has written.

• Thinking with the Body and the World, Barbara Tversky: To me this felt like biggest literature review of existence of geometry, patterns, etc in human evolution and then connecting the dot throughout the history. This blog provides good overview of what the talk was about