Research map / 3D vision / Depth and stereo
Neural Light Field Estimation for Street Scenes with Differentiable Virtual Object Insertion
Zian Wang, Wenzheng Chen, David Acuna, Jan Kautz, Sanja Fidler
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 positional and peak encoding to sky encoder and decoder (ablation) | sky modeling network · HDR sky reconstruction | HDR MSE (x1e-2): 8.31 -> 7.49 | Improved |
| add positional and peak encoding to sky encoder and decoder (ablation) | sky modeling network · HDR sky reconstruction | peak direction median angular error: 4.09° -> 3.38° | Improved |
| add explicit peak direction and intensity to the sky vector (ablation) | sky modeling network · HDR sky reconstruction | HDR MSE (x1e-2): 11.01 -> 7.49 | Improved |
| add explicit peak direction and intensity to the sky vector (ablation) | sky modeling network · HDR sky reconstruction | peak direction median angular error: 7.93° -> 3.38° | Improved |
| use explicit peak direction and intensity instead of latent code with explicit azimuth | sky modeling network · HDR sky reconstruction | HDR MSE (x1e-2): 10.56 -> 7.49 | Improved |
| use hybrid lighting instead of baselines for sphere insertion | hybrid lighting model · diffuse sphere insertion | rendering MSE (x1e-3): 2.30 -> 1.79 | Improved |
| use hybrid lighting instead of baselines for sphere insertion | hybrid lighting model · specular sphere insertion | rendering MSE (x1e-3): 3.13 -> 2.41 | Improved |
| add six surrounding camera views as input (ablation) | lighting prediction · HoliCity | sun location median angular error: 22.43° -> 19.91° | Improved |
| add six surrounding camera views as input (ablation) | lighting prediction · nuScenes | PSNR: 14.49 -> 17.96 | Improved |
| add six surrounding camera views as input (ablation) | lighting prediction · nuScenes | si-PSNR: 15.35 -> 18.45 | Improved |
| add inserted virtual objects without lighting estimation as augmentation | 3D object detector · nuScenes 3D detection (10%) | mAP: 0.190 -> 0.201 | Improved |
| add objects rendered with estimated lighting instead of fixed dome lighting | 3D object detector · nuScenes 3D detection (10%) | mAP: 0.201 -> 0.211 | Improved |
Closest papers by meaning
- Estimating Spatially-Varying Lighting in Urban Scenes with Disentangled Representation (ECCV 2022 · similarity 0.84)
- Spatio-Temporal Outdoor Lighting Aggregation on Image Sequences Using Transformer Networks (IJCV 2022 · similarity 0.84)
- Deep Parametric Indoor Lighting Estimation (ICCV 2019 · similarity 0.84)
- Spatially-Varying Outdoor Lighting Estimation From Intrinsics (CVPR 2021 · similarity 0.83)
- Neural Fields Meet Explicit Geometric Representations for Inverse Rendering of Urban Scenes (CVPR 2023 · similarity 0.83)
- Multi-Level Neural Scene Graphs for Dynamic Urban Environments (CVPR 2024 · similarity 0.81)
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.
