CVPR Daily - Thursday

When I started, vision and language were very niche areas. We knew all the names in this field. We all knew each other. Now it's grown enormously, like probably all the other areas similarly. Visional language has exploded. It's almost like a field of its own, if you will. Maybe somebody won't like me saying that, but there are just so many different trends and directions within vision and language, within multimodal learning in general. So it has definitely taken off. I think with the arrival of models like CLIP from Open AI, even those who didn't want to now see that there is huge potential in these multi-model models in terms of leveraging from weaker supervision and then bringing this strong knowledge onto a large variety of downstream vision tasks. One good answer generates two questions! What do you find so exciting in the visual language area that you devote the best years of your career to it? And the second question: what is happening now from an innovation and R&D point of view? For the first question, maybe I'll give you a silly answer. I just liked vision and language so much that I wanted to do both, and I couldn't decide which one. Why choose between daddy and mommy? Exactly. I was initially working more in the vision space, but I also had this passion for languages. Then I discovered I can do both. I just couldn't let it go. To the second question, I think really what's driving the progress right now in vision and language, but maybe also other areas, are these huge large-scale, pre-trained foundational models. The models which learn from just enormous amounts of data really began to approach levels of recognition and sometimes even surpass certain expectations that we have. Of course, they are not quote- unquote understanding things yet, but I think we are really getting to the point when things work, and we could do more cool, exciting things building on top of these foundational sorts of models. When I say foundational, I think I should 20 DAILY CVPR Women in Computer Vision Thursday Graduation in Odessa

RkJQdWJsaXNoZXIy NTc3NzU=