Reconstructing occluded humans from monocular video can be nice and fast! Our latest paper "OccFusion: Rendering Occluded Humans with Generative Diffusion Priors" combines 3D Gaussian Splatting with pre-trained 2D diffusion models to render humans behind occlusions. Despite being fast, existing Gaussian-splatting-based human rendering methods like GauHuman render floaters and other artifacts when faced with occlusion. OccFusion accounts for this by incorporating 2D diffusion priors to attain high-quality, artifact-free human renderings. Our method achieves the state-of-the-art in both quantitative results and visual rendering quality despite taking only minutes to train and less than a second to render. This work would not have been possible without the genius of my co-leading author Tiange Xiang and the mentorship and guidance of Scott Delp, Fei-Fei Li, and Ehsan Adeli! For more info, check out our paper and webpage, out now! 📖: https://lnkd.in/gAG5Vkju 🌐: https://lnkd.in/g5CtQtrg
I love your posts, Adam. It's the only intereeting stuff in my LinkedIn feed! This is super cool.
so cool Adam Sun !
CS at Stanford
1mothis is huge!