Research map / 3D vision / Depth and stereo
C4D: 4D Made from 3D through Dual Correspondences
Shizun Wang, Zhenxiang Jiang, Xingyi Yang, Xinchao Wang
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 global alignment with C4D optimization on DUSt3R weights | DUSt3R · Sintel camera pose | RPE rot: 18.038 -> 0.948 | Improved |
| replace global alignment with C4D on MonSt3R weights | MonSt3R · Sintel camera pose | ATE: 0.158 -> 0.103 | Improved |
| replace global alignment with C4D on MASt3R weights | MASt3R · Sintel camera pose | ATE: 0.437 -> 0.448 (ATE); 12.760 -> 1.730 (RPE rot) | Mixed |
| replace global alignment with C4D on static ScanNet | DUSt3R · ScanNet (static) camera pose | ATE: 0.060 -> 0.064 (ATE); 0.751 -> 0.570 (RPE rot) | Mixed |
| replace global alignment with C4D for video depth | DUSt3R · Sintel video depth | Abs Rel: 0.502 -> 0.478 | Improved |
| replace global alignment with C4D for video depth | MASt3R · Sintel video depth | Abs Rel: 0.370 -> 0.379 | Hurt |
| add camera movement alignment (CMA) objective (ablation) | C4D · Sintel camera pose | ATE: 0.140 -> 0.103 | Improved |
| add camera trajectory smoothness (CTS) objective (ablation) | C4D · Sintel camera pose | RPE rot: 1.348 -> 0.705 | Improved |
| add point trajectory smoothness (PTS) objective (ablation) | C4D · Sintel camera pose and depth metrics | pose and depth accuracy: ATE 0.103 -> 0.103; Abs Rel 0.327 -> 0.327 | No clear effect |
| combine CNN encoder with 3D-aware encoder in the point tracker (ablation) | DynPT · TAP-Vid DAVIS | AJ: 42.4 -> 61.6 (vs 3D encoder only); 60.6 -> 61.6 (vs CNN only) | Improved |
| use DynPT-guided motion masks in place of MonST3R's masks | C4D · DAVIS2016 motion segmentation | IoU: 31.57 -> 47.89 | Improved |
| predict mobility in addition to position and occlusion in the tracker | DynPT vs CoTracker · TAP-Vid DAVIS | AJ: 61.8 -> 61.6 | No clear effect |
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.
- MonST3R — MonST3R: A Simple Approach for Estimating Geometry in the Presence of Motion (ICLR 2025) · better on 3 of 3 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
- Stereo4D: Learning How Things Move in 3D from Internet Stereo Videos (CVPR 2025 · similarity 0.90)
- CAT4D: Create Anything in 4D with Multi-View Video Diffusion Models (CVPR 2025 · similarity 0.89)
- Enhancing 3D Reconstruction for Dynamic Scenes (NeurIPS 2025 · similarity 0.88)
- Motion4D: Learning 3D-Consistent Motion and Semantics for 4D Scene Understanding (NeurIPS 2025 · similarity 0.88)
- Flux4D: Flow-based Unsupervised 4D Reconstruction (NeurIPS 2025 · similarity 0.88)
- 4D3R: Motion-Aware Neural Reconstruction and Rendering of Dynamic Scenes from Monocular Videos (NeurIPS 2025 · similarity 0.88)
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.
