Research map / 3D vision / 3D generation
Geometry-guided Feature Learning and Fusion for Indoor Scene Reconstruction
Ruihong Yin, Sezer Karaoglu, Theo Gevers
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 geometry integration (G2FL + G2AFF + C3NL) to NeuralRecon | NeuralRecon · ScanNet | F-score: 0.560 -> 0.619 | Improved |
| add geometry integration to VoRTX | VoRTX · ScanNet | F-score: 0.598 -> 0.630 | Improved |
| add projected surface normals to the Transformer-based feature learning (ablation) | NeuralRecon + G2FL · ScanNet | F-score: 0.566 -> 0.587 | Improved |
| put normal and viewing priors in separate modules rather than the same Transformer (ablation) | G2FL · ScanNet | F-score: 0.590 -> 0.598 | Improved |
| add projected depth to G2FL (ablation) | G2FL · ScanNet | F-score: 0.598 -> 0.602 | Improved |
| use our geometric priors instead of SimpleRecon geometry (ablation) | G2FL · ScanNet | F-score: 0.571 -> 0.602 | Improved |
| add geometry-guided adaptive fusion (G2AFF) on top of G2FL (ablation) | NeuralRecon + G2FL · ScanNet | F-score: 0.602 -> 0.614 | Improved |
| use projective TSDF as occlusion approximation instead of projective occupancy (ablation) | G2AFF · ScanNet | F-score: 0.610 -> 0.609 | Hurt |
| apply 3D normal loss to all voxels without consistency weights or boundary mask (ablation) | NeuralRecon + G2FL + G2AFF · ScanNet | F-score: 0.614 -> 0.609 | Hurt |
| add boundary mask and consistency weight (indicator) to normal loss (ablation) | C3NL · ScanNet | F-score: 0.609 -> 0.619 | Improved |
| use indicator function instead of Gaussian weight in normal loss (ablation) | C3NL · ScanNet | F-score: 0.615 -> 0.619 | Improved |
| add NeRF-like encoding of viewing direction and depth (ablation) | G2FL · ScanNet | F-score: 0.600 -> 0.602 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- NeuralRecon — NeuralRecon: Real-Time Coherent 3D Reconstruction From Monocular Video (CVPR 2021) · better on 4 of 6 shared results
- SimpleRecon — SimpleRecon: 3D Reconstruction without 3D Convolutions (ECCV 2022) · better on 1 of 1 shared result
- TransformerFusion — TransformerFusion: Monocular RGB Scene Reconstruction using Transformers (NeurIPS 2021) · 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
- CVRecon: Rethinking 3D Geometric Feature Learning For Neural Reconstruction (ICCV 2023 · similarity 0.88)
- Neural 3D Scene Reconstruction With Indoor Planar Priors (TPAMI 2024 · similarity 0.88)
- Neural 3D Scene Reconstruction With the Manhattan-World Assumption (CVPR 2022 · similarity 0.87)
- Learning 3D Scene Priors With 2D Supervision (CVPR 2023 · similarity 0.87)
- IndoorGS: Geometric Cues Guided Gaussian Splatting for Indoor Scene Reconstruction (CVPR 2025 · similarity 0.86)
- NC-SDF: Enhancing Indoor Scene Reconstruction Using Neural SDFs with View-Dependent Normal Compensation (CVPR 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.
