Computer Vision News 4 to the model. It can crop out people’s clothes and send them into a new network to extract the features. That is used as conditioning so the model understands color and style preferences. For example, a text prompt to create a person wearing a blue shirt begs the question, what shade of blue? If it is light blue, how light is it? It is not easy to put that into words, but a visual prompt enables a faithful representation. Another exciting aspect of this work is its ability to perform pose and camera view interpolation. Given two pictures, it can interpolate anything in between. This groundbreaking feature, the first of its kind, facilitates the seamless generation of images that transition between poses and camera perspectives. What are the next steps for this work? “Currently, we use a small data set with a plain background,” Soon tells us. “Our next step is to apply that to ICCV Workshop Paper and Demo
RkJQdWJsaXNoZXIy NTc3NzU=