Research map / 3D vision / 3D generation
RodinHD: High-Fidelity 3D Avatar Generation with Diffusion Models
Bowen Zhang, Yiji Cheng, Chunyu Wang, Ting Zhang, Jiaolong Yang, Yansong Tang, Feng Zhao, Dong Chen, Baining Guo
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add task replay (switch avatars more often) when fitting the shared decoder (ablation) | Rodin (512) triplane fitting · Triplane fitting ablation | PSNR: 30.56 -> 30.93 | Improved |
| add identity-aware weight consolidation regularizer (ablation) | Rodin (512) with task replay and weight decay · Triplane fitting ablation | PSNR: 31.00 -> 31.45 | Improved |
| add weight decay (ablation) | Rodin (512) with task replay · Triplane fitting ablation | PSNR: 30.93 -> 31.00 | Improved |
| naively increase triplane resolution to 512 without fixing forgetting (ablation) | Rodin fitting · Triplane fitting ablation | PSNR: 30.31 -> 30.56 | No clear effect |
| naively scale triplane diffusion to 512x512x32 without changing noise schedule (ablation) | Rodin diffusion · Conditional avatar generation | FID: 33.20 -> 35.84 | Hurt |
| add stronger noise schedule for triplane diffusion (ablation) | RodinHD at 512 triplane · Conditional avatar generation | FID: 35.84 -> 27.74 | Improved |
| replace CLIP image encoder with single-scale VAE encoder conditioning (ablation) | RodinHD · Conditional avatar generation | FID: 27.74 -> 27.13 | Improved |
| use multi-scale instead of single-scale VAE feature conditioning (ablation) | RodinHD · Conditional avatar generation | FID: 27.13 -> 26.49 | Improved |
| avoid 2D refinement (keep rendering directly from triplane) | RodinHD vs Rodin* · 3D consistency (NeuS fit) | PSNR: 31.73 -> 35.46 | Improved |
| 2D refiner applied to Rodin output | Rodin · Unconditional avatar generation | FID: 45.70 -> 30.29 | Improved |
| increase triplane channels and resolution (ablation) | triplane fitting of one subject · Triplane res/channel sweep | PSNR: 30.15 -> 31.71 | Improved |
| RodinHD without refiner vs Rodin with 2D refinement on visual fidelity ranking | RodinHD · User study | Visual fidelity rank: 1.26 vs 2.13 | Hurt |
Closest papers by meaning
- CAP4D: Creating Animatable 4D Portrait Avatars with Morphable Multi-View Diffusion Models (CVPR 2025 · similarity 0.89)
- Human-3Diffusion: Realistic Avatar Creation via Explicit 3D Consistent Diffusion Models (NeurIPS 2024 · similarity 0.89)
- Morphable Diffusion: 3D-Consistent Diffusion for Single-image Avatar Creation (CVPR 2024 · similarity 0.89)
- DiffusionGAN3D: Boosting Text-guided 3D Generation and Domain Adaptation by Combining 3D GANs and Diffusion Priors (CVPR 2024 · similarity 0.89)
- DiffusionAvatars: Deferred Diffusion for High-fidelity 3D Head Avatars (CVPR 2024 · similarity 0.89)
- AvatarArtist: Open-Domain 4D Avatarization (CVPR 2025 · similarity 0.89)
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.
