Orest Kupyn

Triangle splats from video diffusion latents yield superior geometric accuracy and visual quality, challenging the dominance of volumetric 3D Gaussians in scene generation.

Multi-modal generation from diffusion transformers can boost semantic segmentation performance by nearly 30% by leveraging temporal perceptual information.

Forget expensive MoCap rigs: PoseDreamer generates synthetic human datasets that outperform real-world data for 3D mesh estimation, thanks to diffusion models and clever sample selection.