Research map / 3D vision / NeRF and radiance fields
Evidential Neural Radiance Fields
Ruxiao Duan, Alex Wong
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 |
|---|---|---|---|
| model pixel color with evidential random mean and variance instead of fixed normal | nerfacto · LF, LLFF, RobustNeRF | PSNR: 28.5538 -> 29.9679 (LF); 17.5281 -> 17.8793 (LLFF); 25.2205 -> 26.2292 (RobustNeRF) | Improved |
| use fixed-variance normal likelihood for uncertainty | nerfacto · Light Field | PSNR: 28.5538 -> 28.0064 | Hurt |
| add Monte Carlo dropout for uncertainty | nerfacto · RobustNeRF | PSNR: 25.2205 -> 24.7693 | Hurt |
| use mixture of Laplace likelihood | nerfacto · RobustNeRF | PSNR: 25.2205 -> 23.7874 | Hurt |
| random mean and variance instead of fixed normal in likelihood | Normal NeRF · LLFF NLL | NLL: 55.3580 -> 0.6765 | Improved |
| use evidential model rather than 5-model ensembles | Evidential NeRF · Training time per 30k steps | minutes: 59.22 -> 13.57 | Improved |
| use evidential model rather than ensembles for sparsification error | Evidential NeRF · LLFF and RobustNeRF | AUSE-RMSE: 0.0513 -> 0.0578 (LLFF); 0.0164 -> 0.0221 (RobustNeRF) | Hurt |
| MoL likelihood versus evidential for NLL | Evidential NeRF · LF NLL | NLL: -2.5393 -> -2.4491 | Hurt |
| increase training images from 10 to 50 (ablation) | Evidential NeRF · RobustNeRF scene | test AU and EU | Mixed |
| select next-best views by epistemic uncertainty instead of random (ablation) | Evidential NeRF · LLFF Horns active learning | test PSNR | Improved |
| suppress density at points with high aleatoric uncertainty | Evidential NeRF · in-the-wild scene | floater removal | Improved |
Closest papers by meaning
- Conditional-Flow NeRF: Accurate 3D Modelling with Reliable Uncertainty Quantification (ECCV 2022 · similarity 0.91)
- Instant Uncertainty Calibration of NeRFs Using a Meta-Calibrator (ECCV 2024 · similarity 0.91)
- Bayes' Rays: Uncertainty Quantification for Neural Radiance Fields (CVPR 2024 · similarity 0.90)
- Neural Visibility Field for Uncertainty-Driven Active Mapping (CVPR 2024 · similarity 0.90)
- ProvNeRF: Modeling per Point Provenance in NeRFs as a Stochastic Field (NeurIPS 2024 · similarity 0.89)
- NeRF On-the-go: Exploiting Uncertainty for Distractor-free NeRFs in the Wild (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.
