Research map / 3D vision / Depth and stereo
SimpleRecon: 3D Reconstruction without 3D Convolutions
Mohamed Sayed, John Gibson, Jamie Watson, Victor Prisacariu, Michael Firman, Clément Godard
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 feature, mask and depth metadata to the cost volume MLP (ablation) | SimpleRecon baseline · ScanNetv2 ablation | Abs Diff: 0.0941 -> 0.0904 | Improved |
| add ray and angle metadata (ablation) | SimpleRecon · ScanNetv2 ablation | Abs Diff: 0.0904 -> 0.0896 | Improved |
| add pose distance metadata (ablation) | SimpleRecon · ScanNetv2 ablation | Abs Diff: 0.0896 -> 0.0885 | Improved |
| add all metadata to the MLP cost volume (ablation) | SimpleRecon baseline · ScanNetv2 mesh eval | F-score: 64.2 -> 67.1 | Improved |
| use 64 feature channels instead of 16 in the dot product cost volume (ablation) | SimpleRecon baseline · ScanNetv2 ablation | Abs Diff: 0.0941 -> 0.0944 | Hurt |
| order keyframes by pose distance instead of shuffling (with metadata) (ablation) | SimpleRecon · ScanNetv2 ablation | Abs Diff: 0.0906 -> 0.0885 | Improved |
| increase source views from 1 (no CV) to 2 (ablation) | SimpleRecon · ScanNetv2 ablation | Abs Diff: 0.1742 -> 0.1230 | Improved |
| increase source views from 2 to 4 (ablation) | SimpleRecon · ScanNetv2 ablation | Abs Diff: 0.1230 -> 0.1036 | Improved |
| increase source views from 4 to 8 (ablation) | SimpleRecon · ScanNetv2 ablation | Abs Diff: 0.1036 -> 0.0885 | Improved |
| use EfficientNetV2 S + R18 backbone instead of MnasNet at 320x256 (ablation) | SimpleRecon · ScanNetv2 ablation | Abs Diff: 0.0947 -> 0.0885 | Improved |
| use SimpleRecon 2D CNN with TSDF fusion instead of 3D CNN update methods | SimpleRecon · Frame integration latency | update latency (ms): 382ms -> 72ms | Improved |
| use SimpleRecon instead of DeepVideoMVS fusion | depth estimation · ScanNetv2 depth | Abs Diff: 0.1186 -> 0.0885 | Improved |
Later papers that compared with this method
9 later accepted papers report SimpleRecon as a baseline in their results tables; 9 of them report a better number on most of the results they share with it.
- DoubleTake: Geometry Guided Depth Estimation (ECCV 2024 · better on 5 of 6 shared results)
- DiffInDScene: Diffusion-based High-Quality 3D Indoor Scene Generation (CVPR 2024 · better on 3 of 3 shared results)
- Deep Cost Ray Fusion for Sparse Depth Video Completion (ECCV 2024 · better on 1 of 1 shared result)
- Virtual Occlusions Through Implicit Depth (CVPR 2023 · better on 3 of 4 shared results)
- FineRecon: Depth-aware Feed-forward Network for Detailed 3D Reconstruction (ICCV 2023 · better on 4 of 4 shared results)
- CVRecon: Rethinking 3D Geometric Feature Learning For Neural Reconstruction (ICCV 2023 · better on 3 of 4 shared results)
- Inner-Outer Aware Reconstruction Model for Monocular 3D Scene Reconstruction (NeurIPS 2023 · better on 2 of 2 shared results)
- DG-Recon: Depth-Guided Neural 3D Scene Reconstruction (ICCV 2023 · better on 1 of 1 shared result)
1 more not listed.
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- NeuralRecon — NeuralRecon: Real-Time Coherent 3D Reconstruction From Monocular Video (CVPR 2021) · better on 2 of 2 shared results
- DELTAS — DELTAS: Depth Estimation by Learning Triangulation And densification of Sparse points (ECCV 2020) · better on 2 of 2 shared results
- DPSNet — DPSNet: End-to-end Deep Plane Sweep Stereo (ICLR 2019) · better on 2 of 2 shared results
- 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
- Neural 3D Scene Reconstruction With the Manhattan-World Assumption (CVPR 2022 · similarity 0.88)
- Associative3D: Volumetric Reconstruction from Sparse Views (ECCV 2020 · similarity 0.87)
- Learning Non-Volumetric Depth Fusion Using Successive Reprojections (CVPR 2019 · similarity 0.87)
- PlaneMVS: 3D Plane Reconstruction From Multi-View Stereo (CVPR 2022 · similarity 0.87)
- 3D Gaussian Flats: Hybrid 2D/3D Photometric Scene Reconstruction (NeurIPS 2025 · similarity 0.86)
- Behind the Veil: Enhanced Indoor 3D Scene Reconstruction with Occluded Surfaces Completion (CVPR 2024 · 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.
