Research map / Video / Video generation
Generative Relightable Avatars
Kunwar Maheep Singh, Christian Theobalt, Rishabh Dabral
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 generative refinement after RelightNet (ablation) | GRA · RHC ablation | LPIPS / FVD: LPIPS 8.92 -> 6.42; FVD 118.24 -> 48.79 | Improved |
| add RelightNet on top of microfacet rendering (ablation) | GRA · RHC ablation | PSNR / LPIPS: PSNR 29.34 -> 30.90; LPIPS 12.26 -> 8.92 | Improved |
| add generative refinement (ablation) | GRA · RHC ablation | PSNR / SSIM: PSNR 30.90 -> 30.49; SSIM 89.25 -> 87.52 | Hurt |
| use atemporal instead of temporal lighting cross-attention (ablation) | GRA · RHC ablation | FVD: 58.18 -> 48.79 | Improved |
| add atemporal lighting conditioning (ablation) | GRA · RHC ablation | FVD: 58.21 -> 48.79 | Improved |
| add VACE control (ablation) | GRA · RHC ablation | LPIPS / FVD: LPIPS 6.93 -> 6.42; FVD 61.84 -> 48.79 | Improved |
| add RelightNet before generative refinement (ablation) | GRA · RHC ablation | PSNR / LPIPS: PSNR 29.56 -> 30.49; LPIPS 6.66 -> 6.42 | Improved |
| finetune VACE video model (ablation) | GRA · RHC ablation | FVD: 230.94 -> 48.79 | Improved |
| use pretrained video prior (ablation) | GRA · RHC ablation | FVD: 394.91 -> 48.79 | Improved |
| use video model instead of image relighting model (Neural Gaffer) (ablation) | GRA · RHC ablation | FVD: 102.10 -> 48.79 | Improved |
| full per-subject finetuning vs few-shot adaptation (ablation) | GRA · RHC ablation | LPIPS: 7.01 -> 6.42 | Improved |
| use 4-view tracking instead of 2-view tracking (ablation) | GRA · tracking ablation | LPIPS: 11.23 -> 7.12 | Improved |
Closest papers by meaning
- Relightable and Animatable Neural Avatars from Videos (AAAI 2024 · similarity 0.90)
- BecomingLit: Relightable Gaussian Avatars with Hybrid Neural Shading (NeurIPS 2025 · similarity 0.89)
- Relightable Gaussian Codec Avatars (CVPR 2024 · similarity 0.89)
- NECA: Neural Customizable Human Avatar (CVPR 2024 · similarity 0.89)
- IntrinsicAvatar: Physically Based Inverse Rendering of Dynamic Humans from Monocular Videos via Explicit Ray Tracing (CVPR 2024 · similarity 0.89)
- Vid2Avatar-Pro: Authentic Avatar from Videos in the Wild via Universal Prior (CVPR 2025 · 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.
