Research map / 3D vision / NeRF and radiance fields
Learning to Factorize and Relight a City
Andrew Liu, Shiry Ginosar, Tinghui Zhou, Alexei A. Efros, Noah Snavely
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 bi-color shading instead of mono-color shading (ablation) | Full model · Scene consistency (GSV-TM) | MSE: 0.077 -> 0.071 | Improved |
| add bi-color shading instead of mono-color shading (ablation) | Full model · Space-time completion (GSV-TM) | MSE: 0.215 -> 0.196 | Improved |
| add alignment training (ablation) | Full model · Scene consistency (GSV-TM) | MSE: 0.082 -> 0.071 | Improved |
| add alignment training (ablation) | Full model · Space-time completion (GSV-TM) | MSE: 0.201 -> 0.196 | No clear effect |
| align test stacks at test time (ablation) | Full model · Scene consistency (GSV-TM) | MSE: 0.090 -> 0.071 | Improved |
| align test stacks at test time (ablation) | Full model · Space-time completion (GSV-TM) | MSE: 0.210 -> 0.196 | Improved |
| add azimuth encoder with sun normalization (ablation) | Full model · Scene consistency (GSV-TM) | MSE: 0.072 -> 0.071 | No clear effect |
| add azimuth encoder with sun normalization (ablation) | Full model · Space-time completion (GSV-TM) | MSE: 0.240 -> 0.196 | Improved |
| replace unsupervised azimuth with supervised azimuth encoder | Azimuth encoder · Sun azimuth (Test-GSV) | cosine similarity: 0.806 -> 0.864 | Improved |
| replace unsupervised azimuth with supervised azimuth encoder | Azimuth encoder · Sun azimuth (Laval) | cosine similarity: 0.771 -> 0.831 | Improved |
| use learned factorization instead of MLE Intrinsics | Intrinsic decomposition · Scene consistency (GSV-TM) | MSE: 0.114 -> 0.071 | Improved |
| use learned factorization instead of BigTime | Intrinsic decomposition · Scene consistency (GSV-TM) | MSE: 0.180 -> 0.071 | Improved |
Closest papers by meaning
- Estimating Spatially-Varying Lighting in Urban Scenes with Disentangled Representation (ECCV 2022 · similarity 0.88)
- SUDS: Scalable Urban Dynamic Scenes (CVPR 2023 · similarity 0.86)
- Multi-Level Neural Scene Graphs for Dynamic Urban Environments (CVPR 2024 · similarity 0.85)
- EMPLACE: Self-Supervised Urban Scene Change Detection (AAAI 2025 · similarity 0.84)
- 3D StreetUnveiler with Semantic-aware 2DGS - a simple baseline (ICLR 2025 · similarity 0.84)
- Dynamic 3D Gaussian Fields for Urban Areas (NeurIPS 2024 · similarity 0.83)
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.
