CVPR Daily - Wednesday

So you were in a hurry to make it? We were in such a hurry to make it out! Due to the fact that we see other projects during the project, we will report numbers like performances and benchmark performance of models. So we try so hard to beat all of them through careful model training techniques. Also, we are trying to cover all the benchmarks that we can think of in vision language research. Some of the papers around the same time that were archived focused on three or four benchmarks, and the other one focused on another three or four. So we're trying to cover all of the benchmarks. Then after our paper was out, and we successfully published it at ECCV 2020, we did tons of follow-up work. So that also is an important reason to get high- cited works. Not only do you need to promote your work to other people so they recognize your work and cite your work, but you need to do follow-up work yourself. Basically, it's another way of promoting your work. It's like a progressive way to accumulate the citation. From a level of zero to 100. How much do you enjoy this game? 100. [ laughs ] I hope you have many more hundreds in the future. I have one last question for you. You come from a very special place, from far away. You evolve in a community that is changing dramatically. What is your point of view? What would you like to change in this community? That's a very hard question. Thank you. I came from far away to ask this question. [ both laugh ] On the research side, I think it's always good to bring up more diversity, no matter the background of where the research is coming from. Also, I feel like the diversity in culture will somewhat reflect in the diversity of people's research ideas as well. I somewhat saw this in other people, like my professor's work studying the social bias of the language models. Sometimes you think it's not an issue in your cultural background, but then in other people's cultural background, they will see the social bias. People are just so used to the social norm. Nowadays, large language models or large multimodal models that are coming out, like GPT-4, will be deployed to products like Microsoft products. Studying social bias in these large models is very important. I think we should change to maybe accepting PhD students or giving work offers to researchers to bring more diversity into the community. If you look around, you will see there are more Asian faces. I see quite a few Asians here at CVPR … Yeah, a lot, right? Asian faces in this community. I think we need to pay attention to bringing more minorities into the field as well. 27 DAILY CVPR Wednesday Linjie Li