COSMI: Compositional Synthesis of Multi-Object Interactions
Preprint
COSMI composes single-object captures into a dataset of multi-object interactions and trains a single diffusion model on it that generates interactions with one to five objects.
daniel.eskandar@student.uni-tuebingen.de
I am a Master’s student in Machine Learning at the University of Tübingen, in the Real Virtual Humans group led by Prof. Dr. Gerard Pons-Moll. I am interested in 3D Computer Vision and Virtual Humans, with a focus on clothing, human–object interaction, and head avatars. I am supported by the Zuse School ELIZA Research-Oriented Master’s Scholarship.
Preprint
COSMI composes single-object captures into a dataset of multi-object interactions and trains a single diffusion model on it that generates interactions with one to five objects.
PhysHuman Workshop @ CVPR 2026
Oral PresentationDAMA introduces a layer-aware Gaussian representation for simulation-ready, compositional avatars.
Lightweight visualizer for Gaussian head avatars compatible with PhysHead (CVPR 2026), enabling interactive editing with FLAME-driven control, hair dynamics, coloring, cutting, and hairstyle transfer.