ECCV 2020 Daily - Wednesday

3 Serge Belongie 9 recognitionbeforewestartedthisproject was focused on basic-level categories. Basic-level categories are things like bicycleversus cat andareobviouslyavery important stepping stone – computer vision absolutely needed to tackle that problem – but I discovered that this was very far removed from what would impress a mom-and-pop user. If you show somebody who is not an expert in this area a mobile phone, point it at a bottle and it says bottle, to a computer vision person that is amazing because they have a sense of what was necessary to do that. To a non-expert that just seems obvious. I found it frustrating that as a field we were not telling users things they did not already know. Pursuing this work was partly borne out of frustration that our field was not asking the right questions. Like if you were a scientist who sent people to the moon, the public would say, “How extraordinary!” but classifying butterflies and teaching a computer what a butterfly is is just as difficult, but will never have the same recognition. Yes, exactly. I remember when I was in grad school, I happened to be sharing an office with computer graphics researchers. This is between 1995 and 2000. At the time, computer graphicswas on top of the research world. SIGGRAPH was a very famous conference. There were a lot of very flashy Hollywood- connected companies and computer Student Zekun Hao Student Menglin Jia DAILY W e d n e s d a y