Research map / 3D vision / Face and shape
D-Rex : Diffusion Rendering for Relightable Expressive Avatars
Timo Teufel, xilong zhou, Umar Iqbal, Jan Kautz, Marc Habermann, Vladislav Golyanik, Christian Theobalt
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 |
|---|---|---|---|
| relight with fine-tuned video diffusion vs physically-based MeshAvatar | MeshAvatar · Novel view/motion/light | PSNR: 22.38 -> 26.30 | Improved |
| replace MeshAvatar driving avatar with EVA (ablation) | D-Rex · Novel view/motion/light | PSNR: 25.73 -> 26.30 | Improved |
| zero-shot IC-Light relighting | EVA · Novel view/motion/light | PSNR: 26.30 -> 10.41 | Hurt |
| fine-tune diffusion relighting model (ablation) | DiffusionRenderer · Subjects 2,3 | PSNR: 19.18 -> 27.62 | Improved |
| start from pretrained prior vs random init (ablation) | DiffusionRenderer full fine-tune · Subjects 2,3 | PSNR: 21.27 -> 27.88 | Improved |
| LoRA vs full fine-tuning (ablation) | DiffusionRenderer · Subjects 2,3 | PSNR: 27.88 (full) vs 27.62 (LoRA) | No clear effect |
| mask background in training pairs (ablation) | D-Rex · 4 subjects | PSNR: 26.30 -> 26.82 | Improved |
| train on EVA renderings (enhancement) rather than flat-lit frames (ablation) | D-Rex · 4 subjects | LPIPS: 0.077 -> 0.072; PSNR 26.3 -> 26.12 | Mixed |
Closest papers by meaning
- BecomingLit: Relightable Gaussian Avatars with Hybrid Neural Shading (NeurIPS 2025 · similarity 0.89)
- Morphable Diffusion: 3D-Consistent Diffusion for Single-image Avatar Creation (CVPR 2024 · similarity 0.89)
- Relightable Gaussian Codec Avatars (CVPR 2024 · similarity 0.88)
- SynthLight: Portrait Relighting with Diffusion Model by Learning to Re-render Synthetic Faces (CVPR 2025 · similarity 0.88)
- IntrinsicAvatar: Physically Based Inverse Rendering of Dynamic Humans from Monocular Videos via Explicit Ray Tracing (CVPR 2024 · similarity 0.88)
- Relightable and Animatable Neural Avatars from Videos (AAAI 2024 · similarity 0.88)
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.
