Research map / 3D vision / NeRF and radiance fields
Direct Voxel Grid Optimization: Super-Fast Convergence for Radiance Fields Reconstruction
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 256³ rather than 160³ fine-grid voxels (ablation) | DirectVoxGO · Synthetic-NeRF, Synthetic-NSVF, BlendedMVS, Tanks and Temples | PSNR: 31.95 -> 32.80; 35.08 -> 36.21; 28.02 -> 28.64; 28.41 -> 28.82 | Helped |
| Use post-activation instead of pre-activation interpolation (ablation) | DirectVoxGO · Synthetic-NeRF, Synthetic-NSVF, BlendedMVS, Tanks and Temples | PSNR: 30.84 -> 31.39; 32.66 -> 35.08; 27.39 -> 27.97; 27.44 -> 27.66 | Helped |
| Use post-activation instead of in-activation interpolation (ablation) | DirectVoxGO · Synthetic-NeRF, Synthetic-NSVF, BlendedMVS, Tanks and Temples | PSNR: 29.91 -> 31.39; 32.42 -> 35.08; 27.29 -> 27.97; 27.52 -> 27.66 | Helped |
| Use low-density initialization (ablation) | DirectVoxGO coarse geometry optimization · Synthetic-NeRF, Synthetic-NSVF, BlendedMVS, Tanks and Temples | PSNR: 28.88 -> 31.39; 25.12 -> 35.08; 22.17 -> 27.97; 25.33 -> 27.66 | Helped |
| Use view-count-based learning rates (ablation) | DirectVoxGO coarse geometry optimization · Synthetic-NeRF, Synthetic-NSVF, BlendedMVS, Tanks and Temples | PSNR: 31.40 -> 31.39; 35.03 -> 35.08; 27.37 -> 27.97; 27.59 -> 27.66 | Helped |
| Increase the fine grid from 100³ to 160³ voxels (ablation) | DirectVoxGO · Robot | PSNR: 32.65 -> 35.54 | Helped |
| Increase the coarse grid from 64³ to 100³ voxels (ablation) | DirectVoxGO · Robot | PSNR: 34.91 -> 35.54 | Helped |
| Reduce training sampling step from 1.00 to 0.50 voxel (ablation) | DirectVoxGO · Robot | PSNR: 34.17 -> 35.54 | Helped |
| Use a finer test-time sampling step, 0.25 rather than 0.50 voxel (ablation) | DirectVoxGO · Robot | PSNR: 35.54 -> 35.72 | Helped |
| Progressively scale fine-stage voxel grids (ablation) | DirectVoxGO · Robot | PSNR: 35.50 -> 35.54 | Helped |
| Apply both coarse- and fine-grid free-space skipping (ablation) | DirectVoxGO fine stage · Robot | training minutes: 15.6 -> 7.3 | Helped |
| Apply coarse-grid free-space skipping rather than fine-grid skipping alone (ablation) | DirectVoxGO fine stage · Robot | PSNR: 34.89 -> 35.54 | Helped |
Closest papers by meaning
- NeRFLight: Fast and Light Neural Radiance Fields Using a Shared Feature Grid (CVPR 2023 · similarity 0.93)
- Sparse Voxels Rasterization: Real-time High-fidelity Radiance Field Rendering (CVPR 2025 · similarity 0.93)
- PyNeRF: Pyramidal Neural Radiance Fields (NeurIPS 2023 · similarity 0.93)
- Learning with Unreliability: Fast Few-shot Voxel Radiance Fields with Relative Geometric Consistency (CVPR 2024 · similarity 0.91)
- CVT-xRF: Contrastive In-Voxel Transformer for 3D Consistent Radiance Fields from Sparse Inputs (CVPR 2024 · similarity 0.91)
- FrugalNeRF: Fast Convergence for Extreme Few-shot Novel View Synthesis without Learned Priors (CVPR 2025 · similarity 0.91)
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.