Research map / 3D vision / Depth and stereo
R3D3: Dense 3D Reconstruction of Dynamic Scenes from Multiple Cameras
Aron Schmied, Tobias Fischer, Martin Danelljan, Marc Pollefeys, Fisher Yu
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 multi-camera DBA to geometric estimation (ablation) | R3D3 geometric estimation · DDAD | ATE** (m): 2.134 -> 0.435 | Improved |
| add VKITTI fine-tuning of geometric estimation (ablation) | R3D3 geometric estimation · DDAD | Abs Rel: 0.442 -> 0.320 | Improved |
| add VKITTI fine-tuning (ablation) | R3D3 geometric estimation · DDAD | ATE (m, absolute scale): 1.672 -> 2.356 | Hurt |
| add depth refinement network on top of geometric estimation (ablation) | R3D3 · DDAD | Abs Rel: 0.320 -> 0.162 | Improved |
| add geometric estimation prior to refinement network (ablation) | R3D3 · DDAD | delta_1.25: 0.715 -> 0.811 | Improved |
| add confidence weights as refinement network input (ablation) | depth refinement network · DDAD | Abs Rel: 0.181 -> 0.173 | Improved |
| sparsify input depth by w_i < beta in addition to confidence (ablation) | depth refinement network · DDAD | Abs Rel: 0.173 -> 0.162 | Improved |
| use new co-visibility graph construction instead of original algorithm (ablation) | R3D3 · DDAD | inference time (s): 3.23 -> 0.35 | Improved |
| new co-visibility graph versus original on pose accuracy (ablation) | R3D3 · DDAD | ATE (m): 0.982 -> 1.235 | Hurt |
| full R3D3 versus single-camera multi-frame methods on front camera | R3D3 · DDAD front camera | Abs Rel: 0.135 -> 0.128 (Abs Rel); RMSE 12.477 vs 13.214 | Mixed |
| combine geometric estimation with refinement; effect on side-view cameras (ablation) | R3D3 · DDAD per-camera | Abs Rel (Front Left): 0.213 -> 0.160 (refinement only vs full) | Improved |
Later papers that compared with this method
3 later accepted papers report R3D3 as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- GaussianOcc: Fully Self-supervised and Efficient 3D Occupancy Estimation with Gaussian Splatting (ICCV 2025 · better on 1 of 1 shared result)
- SS3DM: Benchmarking Street-View Surface Reconstruction with a Synthetic 3D Mesh Dataset (NeurIPS 2024 · better on 4 of 4 shared results)
- M^2Depth: Self-supervised Two-Frame Multi-camera Metric Depth Estimation (ECCV 2024 · better on 1 of 1 shared result)
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.
- DROID-SLAM — DROID-SLAM: Deep Visual SLAM for Monocular, Stereo, and RGB-D Cameras (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
- Dense Dynamic Scene Reconstruction and Camera Pose Estimation from Multi-View Videos (ECCV 2026 · similarity 0.91)
- Stereo4D: Learning How Things Move in 3D from Internet Stereo Videos (CVPR 2025 · similarity 0.89)
- Layered Motion Fusion: Lifting Motion Segmentation to 3D in Egocentric Videos (CVPR 2025 · similarity 0.88)
- DUSt3R: Geometric 3D Vision Made Easy (CVPR 2024 · similarity 0.88)
- 4D3R: Motion-Aware Neural Reconstruction and Rendering of Dynamic Scenes from Monocular Videos (NeurIPS 2025 · similarity 0.87)
- MUSt3R: Multi-view Network for Stereo 3D Reconstruction (CVPR 2025 · 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.
