Research map / 3D vision / Face and shape
Sphere-Guided Training of Neural Implicit Surfaces
Read the paperCheck 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 |
|---|---|---|---|
| Add sphere-guided training | NeuS · Realistic Synthetic 360 | Mean Chamfer distance: 0.70 -> 0.59 | Helped |
| Add sphere-guided training | NeuS · DTU | Mean Chamfer distance: 0.78 -> 0.72 | Helped |
| Add sphere-guided training | NeuralWarp · DTU | Mean Chamfer distance: 0.68 -> 0.68 | No clear effect |
| Add sphere-guided training | VolSDF* · DTU | Mean Chamfer distance: 0.94 -> 0.77 | Helped |
| Add sphere-guided training | NeuS · Realistic Synthetic 360 | PSNR: 29.26 -> 30.52 | Helped |
| Add sphere-guided training | NeuS · Realistic Synthetic 360 | SSIM: 0.940 -> 0.951 | Helped |
| Add sphere-guided training | NeuS · Realistic Synthetic 360 | LPIPS: 0.075 -> 0.063 | Helped |
| Replace gradient-trained sphere cloud with NSVF-style sparse voxel octree (ablation) | NeuS sphere-guided training · Realistic Synthetic 360 | Chamfer distance: Chair: 0.39 -> 0.37; Drums: 1.20 -> 3.12; Ficus: 0.40 -> 0.49; Hotdog: 0.57 -> 0.52; Lego: 0.61 -> 0.64; Mats: 0.31 -> 0 | Hurt |
| Add sphere repulsion loss (ablation) | NeuS sphere-guided training · Realistic Synthetic 360 | Chamfer distance: Chair: 0.76 -> 0.39; Drums: 2.37 -> 1.20; Ficus: 0.79 -> 0.40; Hotdog: 2.71 -> 0.57; Lego: 1.03 -> 0.61; Mats: 0.95 -> 0 | Helped |
| Add sphere-guided ray sampling (ablation) | NeuS sphere-guided training · Realistic Synthetic 360 | Chamfer distance: Chair: 0.36 -> 0.39; Drums: 1.79 -> 1.20; Ficus: 0.49 -> 0.40; Hotdog: 0.55 -> 0.57; Lego: 0.69 -> 0.61; Mats: 0.43 -> 0 | Mixed |
| Add sphere-guided ray marching (ablation) | NeuS sphere-guided training · Realistic Synthetic 360 | Chamfer distance: Chair: 0.38 -> 0.39; Drums: 0.81 -> 1.20; Ficus: 0.40 -> 0.40; Hotdog: 0.54 -> 0.57; Lego: 0.61 -> 0.61; Mats: 0.29 -> 0 | Mixed |
| Add sphere-guided ray marching (ablation) | NeuS with 64 points per ray · Realistic Synthetic 360 | Chamfer distance: Chair: 0.38 -> 0.40; Drums: 1.64 -> 0.79; Ficus: 0.52 -> 0.45; Hotdog: 0.55 -> 0.59; Lego: 0.65 -> 0.60; Mats: 0.28 -> 0 | Mixed |
Closest papers by meaning
- Learning to Infer Implicit Surfaces without 3D Supervision (NeurIPS 2019 · similarity 0.89)
- NeAT: Learning Neural Implicit Surfaces With Arbitrary Topologies From Multi-View Images (CVPR 2023 · similarity 0.89)
- RayDF: Neural Ray-surface Distance Fields with Multi-view Consistency (NeurIPS 2023 · similarity 0.89)
- Iso-Points: Optimizing Neural Implicit Surfaces With Hybrid Representations (CVPR 2021 · similarity 0.88)
- Differentiable Neural Surface Refinement for Modeling Transparent Objects (CVPR 2024 · similarity 0.88)
- NeuSurf: On-Surface Priors for Neural Surface Reconstruction from Sparse Input Views (AAAI 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.