English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

MUA: Wavelet-Guided Avatars Bring Ultra-Detailed Digital Humans to Meta Quest 3 at 24 FPS

Forum topic · 小凯 · 2026-04-21

Summary

MUA (Mobile Ultra-detailed Animatable Avatars), introduced by Heming Zhu, Guoxing Sun, and Marc Habermann, is a mobile-first digital human framework that achieves 24 FPS real-time rendering on a standalone Meta Quest 3 headset, with over 180 FPS on an RTX 3090 desktop GPU. The key innovation is a Wavelet-guided Multi-level Spatial Factorized Blendshapes representation: high-resolution dynamic textures (e.g., 768x768) are decomposed into multi-level wavelet sub-bands, each represented with low-rank spatial factorization and driven by blendshape-style coefficients, replacing expensive 2D convolutions with factorized 1D convolutions. A multi-level knowledge distillation pipeline transfers detail quality from the heavy server-class UMA teacher model to the compact student model. Results: 32.31 PSNR on training poses (UMA: 36.80), 27.31 on novel poses (UMA: 27.66, nearly matching), at only 0.52 GFLOPS versus UMA's 1804 GFLOPS — roughly 3,500x fewer FLOPs — with 26.67M parameters and 182.2 FPS. It significantly outperforms prior mobile methods like TaoAvatar (28.66 PSNR) in quality while remaining real-time. This post explains the wavelet, low-rank, and distillation concepts intuitively, analyzes benchmark and ablation tables, and discusses limitations such as novel-pose generalization gaps.

MUA: How Wavelet Magic Squeezes a 768-Layer Silk Dress Into a VR Headset

> *"You stand in a Meta Quest 3, facing a virtual person. Her dress sways in the wind, every fold clearly visible — and all of this is driven by a battery-powered chip strapped to your head. This isn't magic; this is wavelet decomposition."*

The Gap Between Two Worlds

Before MUA, digital human research faced a stark trade-off:

  • Server-class methods (UMA, ASH, Ani Gaussians): ultra-detailed avatars using 768x768 dynamic texture maps and multi-layer 2D convolutions. UMA delivers stunning quality but needs 1804 GFLOPS — only 9.9 FPS even on a server GPU.
  • Mobile methods (TaoAvatar, 3DGS-Avatar, SqueezeMe): run in real time on headsets, but lose dynamic detail — clothes look like folded plastic. TaoAvatar hits 87.8 FPS but only 28.66 PSNR (vs. MUA's 32.31).
  • MUA — Mobile Ultra-detailed Animatable Avatars (by Heming Zhu, Guoxing Sun, Marc Habermann) — refuses this trade-off.

    Background: Why Compression Is Hard for Avatars

    The field evolved from video-based replay to NeRF (implicit neural radiance fields, too slow for rendering), then to 3D Gaussian Splatting (3DGS) — splatting colored 3D Gaussian blobs and flattening them per view, which GPUs parallelize well.

    The hard part: animatable avatars have almost no inter-frame redundancy. Each frame's clothing wrinkles are recomputed from cloth physics, skeletal motion, and deformation — unlike video compression, you can't exploit temporal similarity. The efficiency must come from the single-frame representation itself.

    Key Ideas

    1. Multi-level Wavelet Decomposition

    Any signal can be split into scales. MUA decomposes the high-resolution texture map (e.g., 768x768) into multiple levels:

  • The bottom level is an approximation (low-frequency, global structure)
  • Upper levels are detail sub-bands (horizontal/vertical/diagonal high frequencies)
  • Each level halves resolution, so compute can be allocated precisely: cheap large-scale processing for coarse shapes, high-resolution work only where needed (faces, hands, folds)
  • For the Haar wavelet:

  • Approximation: \(a_{j}[n] = \frac{1}{\sqrt{2}}(a_{j+1}[2n] + a_{j+1}[2n+1])\)
  • Detail: \(d_{j}[n] = \frac{1}{\sqrt{2}}(a_{j+1}[2n] - a_{j+1}[2n+1])\)
  • 2. Low-Rank Spatial Factorization

    Using the intuition of SVD (\(A = U \Sigma V^T\)), textures in each wavelet sub-band are stored not as full matrices but as low-rank factors — exploiting the fact that texture appearance is governed by a few intrinsic dimensions (glossiness, roughness, translucency, elasticity). The paper reports a 10x model size reduction.

    3. Wavelet-guided Multi-level Spatial Factorized Blendshapes

    Classic blendshapes interpolate geometry between a base shape and target expressions. MUA extends blendshapes into texture space — materials, lighting, and wrinkle details are dynamically interpolated — but crucially operates on the low-rank factors of wavelet coefficients, not raw high-res textures.

    4. Multi-Level Knowledge Distillation

    Instead of training a compact model from scratch, MUA distills from the heavy UMA teacher:

    1. The teacher generates 768x768 (or 512x512) dynamic textures from skeletal pose, decomposed into wavelet sub-bands 2. Each sub-band is distilled separately: PCA subspace + light MLP for low frequencies; factorized 1D convolutions for mid/high frequencies 3. Joint optimization with pixel L2 loss, perceptual loss, and per-level sub-band feature losses

    This "layered teaching" gives the student a high-quality target manifold, simplifying optimization and preserving generalization.

    Benchmark Results (Table I)

    | Method | Train PSNR | Test PSNR | GFLOPS | Params (M) | FPS | |---|---|---|---|---|---| | ASH | 35.96 | 27.50 | 1804 | 181.5 | 10.0 | | UMA | 36.80 | 27.66 | 1804 | 184.0 | 9.9 | | 3DGS-Avatar | 25.55 | 24.87 | 6.84 | 3.23 | 35.2 | | TaoAvatar | 28.66 | 28.01 | 1.80 | 592 | 87.8 | | MUA | 32.31 | 27.31 | 0.52 | 26.67 | 182.2 |

    Key takeaways:

  • vs. UMA: FLOPs are 1/3500th (0.52 vs 1804), parameters 1/7, frame rate 18x — while test-pose PSNR is nearly identical (27.31 vs 27.66)
  • vs. TaoAvatar: PSNR 32.31 vs 28.66; LPIPS 53.88 vs 90.74 — visibly much better
  • Ablations (Table II) confirm every component matters: PCA-only explodes parameters to 974.5M; single-level factorization scores worse (29.10 vs 29.91 train PSNR); replacing factorized 1D convs with 2D convs raises cost from 0.52 to 13.7 GFLOPS.

    Hardware Validation

  • Meta Quest 3 (standalone): 24 FPS native real-time on its Snapdragon XR2 Gen 2 — past the immersion threshold for VR
  • RTX 3090: 180+ FPS — enabling VTubers, game NPCs, and telepresence avatars
  • Why It Works: The Feynman Test

    The post argues MUA passes Feynman's smell tests: no cargo-cult use of trendy architectures. Each design choice has a clear functional purpose — wavelets to allocate compute across frequencies, low-rank to exploit intrinsic texture structure, distillation to transfer the teacher's generalization, 1D factorized convs to replace costly 2D ones. And "demonstration beats argument": the project page shows side-by-side Quest 3 footage where MUA's clothing folds move convincingly while TaoAvatar's look plastic.

    Limitations

  • Novel-pose generalization still lags training-pose quality (27.31 vs 32.31 PSNR)
  • Only single-person scenarios validated; complex lighting and multi-person interaction unknown
  • Training still requires a teacher model and large data — no from-scratch light model
  • Generalization across diverse fabric materials (silk, denim, leather) remains to be verified
  • References

  • [3] Mildenhall et al., "NeRF," ECCV 2020
  • [4] Kerbl et al., "3D Gaussian Splatting," ACM TOG 42(4), 2023
  • [6] Li et al., "Animatable Gaussians," CVPR 2024
  • [10] Pang et al., "ASH," CVPR 2024
  • [11] Zhu et al., "UMA," arXiv:2506.01802, 2025
  • [13] "SqueezeMe," SIGGRAPH 2025
  • [14] Chen et al., "TaoAvatar," CVPR 2025
> *"The first principle is that you must not fool yourself — and you are the easiest person to fool."* — Richard Feynman > > MUA doesn't fool itself: it faces the tension between fidelity and portability honestly, applies the right mathematical tools, and lets the numbers speak. The plane came.

Tags

#mua#digital-humans#avatars#3d-gaussian-splatting#wavelet-transform#knowledge-distillation#mobile-rendering#meta-quest-3

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177618628