Research map / 3D vision / NeRF and radiance fields
Representing Volumetric Videos As Dynamic MLP Maps
Sida Peng, Yunzhi Yan, Qing Shuai, Hujun Bao, Xiaowei Zhou
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 dynamic MLP maps instead of a single shared MLP (ablation) | Volumetric video representation · ZJU-MoCap and NHR (two-scene average) | LPIPS: 0.065 -> 0.058 | Improved |
| Add hash tables to feature tri-planes and a single NeRF (ablation) | C-NeRF-style representation · ZJU-MoCap and NHR (two-scene average) | LPIPS: 0.076 -> 0.065 | Improved |
| Increase color MLP map size from 1 × 1 to 4 × 4 (ablation) | Dynamic MLP maps · ZJU-MoCap and NHR (two-scene average) | LPIPS: 0.069 -> 0.065 | Improved |
| Increase color MLP map size from 4 × 4 to 32 × 32 (ablation) | Dynamic MLP maps · ZJU-MoCap and NHR (two-scene average) | LPIPS: 0.065 -> 0.060 | Improved |
| Increase color MLP map size beyond 32 × 32 (ablation) | Dynamic MLP maps · ZJU-MoCap and NHR (two-scene average) | LPIPS | No clear effect |
| Use three orthogonal MLP maps instead of a single XY MLP map (ablation) | Dynamic MLP maps · ZJU-MoCap and NHR (two-scene average) | LPIPS: 0.072 -> 0.058 | Improved |
| Use three orthogonal planes instead of three maps all on XY (ablation) | Dynamic MLP maps · ZJU-MoCap and NHR (two-scene average) | LPIPS: 0.067 -> 0.058 | Improved |
| Add empty space skipping (ablation) | Complete model · ZJU-MoCap and NHR (two-scene average) | Rendering time (milliseconds): 144 -> 24 | Improved |
| Replace dynamic MLP maps with a single shared MLP (ablation) | Volumetric video representation · ZJU-MoCap and NHR (two-scene average) | Rendering time (milliseconds): 24 -> 90 | Hurt |
| Use dynamic MLP maps instead of a single shared MLP with ESS (ablation) | Volumetric video representation · ZJU-MoCap and NHR (two-scene average) | Rendering time (milliseconds): 1345 -> 24 | Improved |
| Use the proposed representation rather than C-NeRF | Volumetric video rendering · NHR | LPIPS: 0.102 -> 0.080 | Improved |
| Use the proposed representation rather than DyNeRF | Volumetric video rendering · NHR | Rendering time (milliseconds): 5195 -> 33 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- D-NeRF — D-NeRF: Neural Radiance Fields for Dynamic Scenes (CVPR 2021) · better on 0 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
- Compact Neural Volumetric Video Representations with Dynamic Codebooks (NeurIPS 2023 · similarity 0.89)
- Neural 3D Video Synthesis From Multi-View Video (CVPR 2022 · similarity 0.87)
- DaReNeRF: Direction-aware Representation for Dynamic Scenes (CVPR 2024 · similarity 0.85)
- OSN: Infinite Representations of Dynamic 3D Scenes from Monocular Videos (ICML 2024 · similarity 0.84)
- Dynamic 3D Gaussian Fields for Urban Areas (NeurIPS 2024 · similarity 0.84)
- VRVVC: Variable-Rate NeRF-Based Volumetric Video Compression (AAAI 2025 · similarity 0.84)
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.
