Research map / 3D vision / Face and shape
Codec Avatar Studio: Paired Human Captures for Complete, Driveable, and Generalizable Avatars
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Releases two paired-capture datasets (Ava-256 with 256 subjects in dome and headset captures; Goliath-4 with 4 subjects in eight modalities) plus code and pretrained models for head, hand, and body avatars.
Problem
Photorealistic avatars need complete (full-body), driveable (headset-driven), and generalizable models, but paired captures of the same subject across capture systems of differing quality are rarely available outside industrial labs.
Approach
Provide open paired-capture datasets (dome scans with headset or phone captures of the same subjects) together with reference implementations of state-of-the-art avatar models, packaged in multiple storage-friendly release sizes using AVIF compression.
Key results
- Ava-256 provides 256 subjects captured with ~170-camera dome (80 cameras in release) and Quest Pro headset IR cameras, with release sizes of 4, 8, 16, and 32 TB.
- Goliath-4 provides 4 subjects captured in eight modalities, totaling 135 minutes of recording per participant on average.
- AVIF was chosen as distribution format for its superior quality-to-space ratio among WEBP, JPEG, JPEG XL, and AVIF on PSNR.
Limitations
- Goliath-4 has only four subjects, limiting research scope.
- Datasets do not capture the long tail of human expressivity such as sweat, tiredness, or blood flow.
- Hard to create multi-identity relightable head decoders with the release.
Experimental setup
Datasets: Ava-256 Goliath-4
Metrics: PSNR
Builds on: Multiface Mixture of Volumetric Primitives Relightable Gaussian Codec Avatars RelightableHands Driving-signal-aware codec avatars
Closest papers by meaning
- RAM-Avatar: Real-time Photo-Realistic Avatar from Monocular Videos with Full-body Control (CVPR 2024 · similarity 0.86)
- Expressive Gaussian Human Avatars from Monocular RGB Video (NeurIPS 2024 · similarity 0.86)
- Vid2Avatar-Pro: Authentic Avatar from Videos in the Wild via Universal Prior (CVPR 2025 · similarity 0.86)
- CapHuman: Capture Your Moments in Parallel Universes (CVPR 2024 · similarity 0.86)
- VLOGGER: Multimodal Diffusion for Embodied Avatar Synthesis (CVPR 2025 · similarity 0.85)
- AvatarGO: Zero-shot 4D Human-Object Interaction Generation and Animation (ICLR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.