Research map / 3D vision / NeRF and radiance fields
Learning to Synthesize a 4D RGBD Light Field From a Single Image
Pratul P. Srinivasan, Tongzhou Wang, Ashwin Sreelal, Ravi Ramamoorthi, Ren Ng
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 |
|---|---|---|---|
| use the factorized depth, Lambertian rendering and occlusion pipeline instead of direct regression of the light field | single-view light field CNN · Flowers test set (all views) | mean L1 error: 0.031 -> 0.026 | Improved |
| add the occlusion/non-Lambertian refinement network to the Lambertian light field (ablation) | Lambertian rendering · Flowers test set (outermost views) | mean L1 error | Improved |
| use depth-based Lambertian rendering instead of appearance flow | light field synthesis · Flowers test set (outermost views) | mean L1 error | Improved |
| use 4 corner input views with light field interpolation instead of a single input view | light field synthesis · Flowers test set (subset of views) | mean L1 error: 0.0176 -> 0.0145 | Hurt |
| switch scene class from flowers to toys | single-view light field synthesis · Toys vs Flowers (all views) | mean L1 error: 0.026 -> 0.027 | No clear effect |
| add the cross-view depth consistency regularization (ablation) | depth estimation network · Flowers test set | ray depth quality | Improved |
Closest papers by meaning
- End-to-end View Synthesis for Light Field Imaging with Pseudo 4DCNN (ECCV 2018 · similarity 0.90)
- Light Field Neural Rendering (CVPR 2022 · similarity 0.90)
- EPINET: A Fully-Convolutional Neural Network Using Epipolar Geometry for Depth From Light Field Images (CVPR 2018 · similarity 0.89)
- Convolutional Networks for Shape From Light Field (CVPR 2016 · similarity 0.88)
- Light Field Networks: Neural Scene Representations with Single-Evaluation Rendering (NeurIPS 2021 · similarity 0.88)
- R2L: Distilling Neural Radiance Field to Neural Light Field for Efficient Novel View Synthesis (ECCV 2022 · 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.
