AHOY! Animatable Humans under Occlusion from YouTube Videos with Gaussian Splatting and Video Diffusion Priors

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• AHOY reconstructs animatable 3D Gaussian avatars from in-the-wild monocular video, even with heavy occlusion.technical 95%
• AHOY utilizes a hallucination-as-supervision pipeline with identity-finetuned diffusion models to generate supervision for unobserved body regions.method 95%
• AHOY employs a two-stage canonical-to-pose-dependent architecture to reconstruct full pose-dependent Gaussian maps from sparse observations.method 95%
• AHOY decouples map-pose and LBS-pose to mitigate multi-view inconsistencies arising from generated data.method 95%
• AHOY uses a head/body split supervision strategy to maintain facial identity during reconstruction.method 95%
• AHOY achieves state-of-the-art reconstruction quality on YouTube videos and multi-view capture data with significant occlusion.result 90%
• The reconstructed avatars from AHOY can be animated with novel poses and composited into 3DGS scenes captured with cell-phone video.result 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card