Research map / 3D vision / 3D generation
GeoRelight: Learning Joint Geometrical Relighting and Reconstruction with Flexible Multi-Modal Diffusion Transformers
Yuxuan Xue, Ruofan Liang, Egor Zakharov, Timur Bagautdinov, Chen Cao, Giljoo Nam, Shunsuke Saito, Gerard Pons-Moll, Javier Romero
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 |
|---|---|---|---|
| jointly generate geometry with the relit image (vs relit image only) (ablation) | GeoRelight · synthetic relighting | PSNR: 21.19 -> 27.49 | Improved |
| jointly generate relit image with geometry versus giving ground-truth geometry (ablation) | GeoRelight · synthetic relighting | PSNR: 26.96 (GT geometry) vs 27.49 (joint) | No clear effect |
| jointly generate normal and geometry with the relit appearance (vs without appearance) (ablation) | GeoRelight · synthetic normal estimation | angular error: 12.24 -> 9.10 | Improved |
| jointly generate point cloud with relit appearance (vs without appearance) (ablation) | GeoRelight · synthetic point cloud | Chamfer distance: 1.00 -> 0.58 | Improved |
| give ground-truth relit image as clear condition for geometry (ablation) | GeoRelight · synthetic normal estimation | angular error: 9.10 -> 8.55 | Improved |
| add in-the-wild data to Synth+Dome mixed training (ablation) | GeoRelight · in-the-wild relighting (qualitative) | lighting balance | Improved |
| add Dome (light stage) data to synthetic-only training (ablation) | GeoRelight · in-the-wild relighting (qualitative) | realism | Mixed |
| use joint multi-modal GeoRelight instead of DiffusionRenderer | relighting · synthetic data | PSNR: 19.28 -> 27.22 | Improved |
| GeoRelight versus NeuralGaffer on HumanOLAT | relighting · HumanOLAT | SSIM: 0.948 (NeuralGaffer) vs 0.935 | Mixed |
| use GeoRelight iNOD joint geometry instead of VGGT/MoGe2 | geometry estimation · synthetic data | Chamfer distance: 3.37 -> 0.766 | Improved |
| use GeoRelight instead of DiffusionRenderer for albedo | albedo estimation · synthetic data | PSNR: 19.46 -> 28.07 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- MoGe2 — MoGe-2: Accurate Monocular Geometry with Metric Scale and Sharp Details (NeurIPS 2025) · better on 3 of 3 shared results
- VGGT — VGGT: Visual Geometry Grounded Transformer (CVPR 2025) · better on 2 of 2 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
- Generative Multiview Relighting for 3D Reconstruction under Extreme Illumination Variation (CVPR 2025 · similarity 0.89)
- SynthLight: Portrait Relighting with Diffusion Model by Learning to Re-render Synthetic Faces (CVPR 2025 · similarity 0.89)
- UniRelight: Learning Joint Decomposition and Synthesis for Video Relighting (NeurIPS 2025 · similarity 0.88)
- Neural Gaffer: Relighting Any Object via Diffusion (NeurIPS 2024 · similarity 0.88)
- IllumiNeRF: 3D Relighting Without Inverse Rendering (NeurIPS 2024 · similarity 0.88)
- Latent Intrinsics Emerge from Training to Relight (NeurIPS 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.
