Research map / 3D vision / Face and shape
CVRecon: Rethinking 3D Geometric Feature Learning For Neural Reconstruction
Ziyue Feng, Liang Yang, Pengsheng Guo, Bing Li
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 |
|---|---|---|---|
| replace back-projected 2D image features with grid-sampled standard cost volume (ablation) | CVRecon on VoRTX · ScanNet2 | F1-Score: 0.703 -> 0.711 | Improved |
| add Camera Ray Compensation to the cost volume (ablation) | CVRecon on VoRTX · ScanNet2 | F1-Score: 0.711 -> 0.728 | Improved |
| add Contextual Compensation by concatenating 2D keyframe feature (ablation) | CVRecon on VoRTX · ScanNet2 | F1-Score: 0.728 -> 0.730 | Improved |
| fuse contextual feature with a single shared 2D conv instead of concat only (ablation) | CVRecon on VoRTX · ScanNet2 | F1-Score: 0.730 -> 0.731 | No clear effect |
| use group convolution with separate weights per depth plane instead of shared conv (ablation) | CVRecon on VoRTX · ScanNet2 | F1-Score: 0.730 -> 0.735 | Improved |
| build ray context from all 7 cost-volume channels instead of one (ablation) | Ray Compensation · ScanNet2 | F1-Score: +0.002 F1 with 7 times memory | No clear effect |
| replace back-projected features with RCCV in Atlas | Atlas · ScanNet2 | F1 Score: 0.596 -> 0.648 | Improved |
| replace back-projection with RCCV in VoRTX (CVRecon) | VoRTX · ScanNet2 (Atlas protocol) | F1 Score @ 5cm: 0.641 -> 0.690 | Improved |
| compare volumetric CVRecon with depth-based SimpleRecon on completeness | CVRecon · ScanNet2 (TransformerFusion protocol) | Completeness: 0.060 (SimpleRecon) vs 0.067 (CVRecon) | 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.
- SimpleRecon — SimpleRecon: 3D Reconstruction without 3D Convolutions (ECCV 2022) · better on 3 of 4 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
- NeuralRecon: Real-Time Coherent 3D Reconstruction From Monocular Video (CVPR 2021 · similarity 0.89)
- GURecon: Learning Detailed 3D Geometric Uncertainties for Neural Surface Reconstruction (AAAI 2025 · similarity 0.88)
- NeuralRecon: Real-Time Coherent 3D Scene Reconstruction From Monocular Video (TPAMI 2024 · similarity 0.88)
- Geometry-guided Feature Learning and Fusion for Indoor Scene Reconstruction (ICCV 2023 · similarity 0.88)
- SparseRecon: Neural Implicit Surface Reconstruction from Sparse Views with Feature and Depth Consistencies (ICCV 2025 · similarity 0.88)
- CoReNet: Coherent 3D Scene Reconstruction from a Single RGB Image (ECCV 2020 · 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.
