Research map / 3D vision / Face and shape
Relighting4D: Neural Relightable Human from Videos
Zhaoxi Chen, Ziwei Liu
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 latent feature psi_t(x) to geometry/reflectance MLPs (ablation) | Relighting4D · BlenderHuman (16 scenes) | PSNR (relighting): 21.1163 -> 26.1475 | Improved |
| add smoothness regularization tau_V, tau_n (ablation) | Relighting4D · BlenderHuman (16 scenes) | PSNR (relighting): 25.3504 -> 26.1475 | Improved |
| add baked geometry supervision (ablation) | Relighting4D · BlenderHuman (16 scenes) | PSNR (relighting): 22.4221 -> 26.1475 | Improved |
| add baked geometry supervision (ablation) | Relighting4D · BlenderHuman (16 scenes) | Normal map error (degrees): 57.0452 -> 32.1803 | Improved |
| add global sparsity prior (minimize H_A) on diffuse map (ablation) | Relighting4D · BlenderHuman (16 scenes) | PSNR (relighting): 27.7545 -> 26.1475 | Mixed |
| add global sparsity prior (minimize H_A) on diffuse map (ablation) | Relighting4D · BlenderHuman (16 scenes) diffuse map | PSNR: 24.0195 -> 28.9517 | Improved |
| add progressive spatial resolution training (ablation) | Relighting4D · BlenderHuman (16 scenes) | PSNR (relighting): 25.5562 -> 26.1475 | Improved |
| use progressive training instead of constant resolution (ablation) | Relighting4D · People-Snapshot | PSNR: 34.56 -> 36.65 | Improved |
| physically based rendering vs NB+LE learnable light embedding | Relighting4D · BlenderHuman (16 scenes) | PSNR (relighting): 22.7957 -> 26.1475 | Improved |
| Relighting4D vs PhySG | relighting · BlenderHuman (16 scenes) | PSNR (relighting): 23.8810 -> 26.1475 | Improved |
Later papers that compared with this method
7 later accepted papers report Relighting4D as a baseline in their results tables; 7 of them report a better number on most of the results they share with it.
- Deep Inverse Shading: Consistent Albedo and Surface Detail Recovery via Generative Refinement (AAAI 2026 · better on 9 of 9 shared results)
- Relightable and Animatable Neural Avatars from Videos (AAAI 2024 · better on 9 of 9 shared results)
- Relightable Neural Actor with Intrinsic Decomposition and Pose Control (ECCV 2024 · better on 3 of 3 shared results)
- Relightable and Animatable Neural Avatar from Sparse-View Video (CVPR 2024 · better on 2 of 2 shared results)
- MeshAvatar: Learning High-quality Triangular Human Avatars from Multi-view Videos (ECCV 2024 · better on 2 of 2 shared results)
- RANA: Relightable Articulated Neural Avatars (ICCV 2023 · better on 4 of 4 shared results)
- Neural Reconstruction of Relightable Human Model from Monocular Video (ICCV 2023 · better on 3 of 3 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Comprehensive Relighting: Generalizable and Consistent Monocular Human Relighting and Harmonization (CVPR 2025 · similarity 0.90)
- UniRelight: Learning Joint Decomposition and Synthesis for Video Relighting (NeurIPS 2025 · similarity 0.88)
- Relightable and Animatable Neural Avatars from Videos (AAAI 2024 · similarity 0.88)
- IllumiNeRF: 3D Relighting Without Inverse Rendering (NeurIPS 2024 · similarity 0.87)
- ROGR: Relightable 3D Objects using Generative Relighting (NeurIPS 2025 · similarity 0.87)
- SwitchLight: Co-design of Physics-driven Architecture and Pre-training Framework for Human Portrait Relighting (CVPR 2024 · similarity 0.87)
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.
