Research map / 3D vision / Depth and stereo
Multi-View Stereo by Temporal Nonparametric Fusion
Yuxin Hou, Juho Kannala, Arno Solin
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 |
|---|---|---|---|
| add GP prior on latent space with Matern pose kernel (ablation) | MVDepthNet-style encoder-decoder · Kernel ablation | L1-rel: 0.1538 -> 0.1298 | Improved |
| add GP prior on latent space with Matern pose kernel (ablation) | MVDepthNet-style encoder-decoder · Kernel ablation | L1: 0.3265 -> 0.2769 | Improved |
| use pose-distance Matern kernel instead of exponential kernel (ablation) | GP-MVS · Kernel ablation | L1-rel: 0.1376 -> 0.1298 | Improved |
| use pose-distance Matern kernel instead of temporal-difference kernel (ablation) | GP-MVS · Kernel ablation | L1-rel: 0.1450 -> 0.1298 | Improved |
| use temporal-difference kernel instead of no GP (ablation) | GP-MVS · Kernel ablation | L1-rel: 0.1538 -> 0.1450 | Improved |
| add GP prior with small translation threshold (ablation) | GP-MVS · SUN3D + 7scenes (t_min=0.1 m) | L1-rel: 0.1474 -> 0.1238 | Improved |
| add GP prior with smaller translation threshold (ablation) | GP-MVS · SUN3D + 7scenes (t_min=0.05 m) | L1-rel: 0.1535 -> 0.1262 | Improved |
| lower translation threshold from 0.1 m to 0.05 m, no GP (ablation) | GP-MVS w/o GP · SUN3D + 7scenes | L1-rel: 0.1474 -> 0.1535 | Hurt |
| lower translation threshold from 0.1 m to 0.05 m, with GP (ablation) | GP-MVS · SUN3D + 7scenes | L1-rel: 0.1238 -> 0.1262 | No clear effect |
| use 5 input frames instead of 2 (ablation) | GP-MVS · 7scenes redkitchen-02 | L1-rel: 0.1615 -> 0.1429 | Improved |
| add GP prior to MVDepthNet with 2 frames (ablation) | MVDepthNet · 7scenes redkitchen-02 (2 frames) | L1-rel: 0.2009 -> 0.1615 | Improved |
| add GP prior (online inference) over MVDepthNet at test time | MVDepthNet · Inference time (desktop) | seconds per frame: 0.066 -> 0.076 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeepMVS — DeepMVS: Learning Multi-View Stereopsis (CVPR 2018) · better on 1 of 1 shared result
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
- DeepVideoMVS: Multi-View Stereo on Video With Recurrent Spatio-Temporal Fusion (CVPR 2021 · similarity 0.89)
- MVS2D: Efficient Multi-View Stereo via Attention-Driven 2D Convolutions (CVPR 2022 · similarity 0.89)
- A Global Depth-Range-Free Multi-View Stereo Transformer Network with Pose Embedding (NeurIPS 2024 · similarity 0.88)
- RIAV-MVS: Recurrent-Indexing an Asymmetric Volume for Multi-View Stereo (CVPR 2023 · similarity 0.88)
- StereoSplat: Metric-Scale Novel View Synthesis via Stereo-Grounded Gaussian Splatting (NeurIPS 2026 · similarity 0.88)
- No Pose, No Problem: Surprisingly Simple 3D Gaussian Splats from Sparse Unposed Images (ICLR 2025 · similarity 0.86)
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.
