Research map / Video / Video generation
DepthSync: Diffusion Guidance-Based Depth Synchronization for Scale- and Geometry-Consistent Video Depth Estimation
Yue-Jiang Dong, Wang Zhao, Jiale Xu, Ying Shan, Song-Hai Zhang
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 scale guidance during denoising (without geometry guidance) (ablation) | DepthCrafter · ScanNet first 35 scenes, 150 frames | AbsRel: 0.174 -> 0.151 | Improved |
| add geometry guidance (without scale guidance) (ablation) | DepthCrafter · ScanNet first 35 scenes, 150 frames | MFC: 0.024 -> 0.018 | Improved |
| combine scale and geometry guidance (ablation) | DepthCrafter · ScanNet first 35 scenes, 150 frames | AbsRel: 0.174 -> 0.137 | Improved |
| apply regularization inside the denoising loop instead of post-optimization (ablation) | DepthSync · ScanNet first 35 scenes, 150 frames | AbsRel: 0.151 -> 0.137 | Improved |
| post geometry optimization alone (after denoising) (ablation) | DepthCrafter · ScanNet first 35 scenes, 150 frames | AbsRel: 0.174 -> 0.175 | No clear effect |
| start guidance earlier (last 5 steps instead of last 2) (ablation) | DepthSync · ScanNet first 25 scenes, 150 frames | AbsRel: 0.137 -> 0.146 | Hurt |
| apply guidance only in the last step (ablation) | DepthSync · ScanNet first 25 scenes, 150 frames | AbsRel: 0.137 -> 0.151 | Hurt |
| start guidance at third-to-last instead of second-to-last step (ablation) | DepthSync · ScanNet first 25 scenes, 150 frames | delta1: 0.801 -> 0.803 | No clear effect |
| apply geometry guidance before scale guidance (ablation) | DepthSync · ScanNet first 25 scenes, 150 frames | AbsRel: 0.145 -> 0.137 | Improved |
| use all four geometry loss terms instead of any single one (ablation) | Geometry guidance · ScanNet first 25 scenes, 150 frames | AbsRel: 0.122 -> 0.108 | Improved |
| apply scale guidance throughout denoising instead of only at initialization | DepthCrafter · Bonn (590 frames) | delta1: 0.801 -> 0.929 | Improved |
| use lightweight scale-only guidance at penultimate step (Ours-S) (ablation) | DepthSync · ScanNet (150 frames) | AbsRel: 0.113 -> 0.128 | Mixed |
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.
- DepthCrafter — DepthCrafter: Generating Consistent Long Depth Sequences for Open-world Videos (CVPR 2025) · better on 8 of 8 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
- Learning Temporally Consistent Video Depth from Video Diffusion Priors (CVPR 2025 · similarity 0.89)
- Video Depth Anything: Consistent Depth Estimation for Super-Long Videos (CVPR 2025 · similarity 0.88)
- GeoVideo: Introducing Geometric Regularization into Video Generation Model (NeurIPS 2025 · similarity 0.87)
- Align3R: Aligned Monocular Depth Estimation for Dynamic Videos (CVPR 2025 · similarity 0.87)
- Depth Any Video with Scalable Synthetic Data (ICLR 2025 · similarity 0.87)
- Seurat: From Moving Points to Depth (CVPR 2025 · 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.
