Research map / 3D vision / Face and shape
DI-Fusion: Online Implicit 3D Reconstruction With Deep Priors
Jiahui Huang, Shi-Sheng Huang, Haoxuan Song, Shi-Min Hu
Check 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 probabilistic SDF uncertainty modeling instead of assuming standard deviation 1.0 (ablation) | DI-Fusion · ICL-NUIM | ATE (cm): Ours (w/o Prob): 1.3, 1.8, 2.6, 15.2; Ours: 1.1, 1.4, 2.6, 6.2 | Improved |
| Add probabilistic SDF uncertainty modeling instead of assuming standard deviation 1.0 (ablation) | DI-Fusion · ICL-NUIM | surface error (cm): Ours (w/o Prob): 0.7, 2.0, 1.2, 4.8; Ours: 0.6, 1.5, 1.1, 4.5 | Improved |
| Use mean rather than max for latent-vector integration (ablation) | DI-Fusion · ICL-NUIM | ATE (cm): Ours (max): 1.4, 2.0, 2.4, 7.1; Ours: 1.1, 1.4, 2.6, 6.2 | Mixed |
| Use mean rather than max for latent-vector integration (ablation) | DI-Fusion · ICL-NUIM | surface error (cm): Ours (max): 0.8, 1.8, 1.2, 4.8; Ours: 0.6, 1.5, 1.1, 4.5 | Improved |
| Use intensity tracking alongside the SDF term at a weight between 500 and 2000 (ablation) | DI-Fusion camera tracking · ICL-NUIM | ATE | Improved |
| Use pure SDF tracking without the intensity term (w=0) (ablation) | DI-Fusion camera tracking · ICL-NUIM sequences with few geometric features | ATE | Hurt |
| Increase intensity-term weight too much (w > 4000) (ablation) | DI-Fusion camera tracking · ICL-NUIM | ATE | Hurt |
| Increase the frame integration interval (ablation) | DI-Fusion geometry integration · ICL-NUIM | ATE | Hurt |
| Integrate too frequently (ablation) | DI-Fusion geometry integration · ICL-NUIM | mapping quality | Hurt |
| Shrink PLIVox size (ablation) | DI-Fusion · 3D reconstruction | reconstruction quality | Improved |
| Shrink PLIVox size (ablation) | DI-Fusion · 3D reconstruction | memory requirement | Hurt |
| Use DI-Fusion rather than TSDF or surfel representations | ICL-NUIM · ICL-NUIM | ATE (cm): DI-Fusion: 1.1, 1.4, 2.6, 6.2; best baseline: 1.7, 1.0, 1.3, 12.5 | Mixed |
Later papers that compared with this method
2 later accepted papers report DI-Fusion as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- CIRCLE: Convolutional Implicit Reconstruction and Completion for Large-Scale Indoor Scene (ECCV 2022 · better on 6 of 7 shared results)
- BNV-Fusion: Dense 3D Reconstruction Using Bi-Level Neural Volume Fusion (CVPR 2022 · better on 2 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
- DIFIX3D+: Improving 3D Reconstructions with Single-Step Diffusion Models (CVPR 2025 · similarity 0.86)
- DITTO: Dual and Integrated Latent Topologies for Implicit 3D Reconstruction (CVPR 2024 · similarity 0.86)
- DiffHuman: Probabilistic Photorealistic 3D Reconstruction of Humans (CVPR 2024 · similarity 0.86)
- MorpheuS: Neural Dynamic 360deg Surface Reconstruction from Monocular RGB-D Video (CVPR 2024 · similarity 0.85)
- Decompositional Neural Scene Reconstruction with Generative Diffusion Prior (CVPR 2025 · similarity 0.85)
- Sparse3D: Distilling Multiview-Consistent Diffusion for Object Reconstruction from Sparse Views (AAAI 2024 · similarity 0.85)
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.
