ZeroComp: Zero-shot Object Compositing from Image Intrinsics via Diffusion ZeroComp is a zero-shot 3D object compositing method that integrates virtual objects into scenes without paired training images, using ControlNet and Stable Diffusion to adjust shading and scene inter...

What better way to kickstart my bluesky account than to advertise for my student's WACV papers! Don't miss ZeroComp, where we use diffusion models as neural renderers that achieve image compositing in a zero-shot fashion: lvsn.github.io/ZeroComp/. With the great @anandbhattad.bsky.social