Research map / Image generation / Super-resolution
DuCos: Duality Constrained Depth Super-Resolution via Foundation Model
Zhiqiang Yan, Zhengxue Wang, Haoye Dong, Jun Li, Jian Yang, Gim Hee Lee
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 depth foundation model prompts to the naive two-branch model (ablation) | naive DSR model · RGB-D-D (real-world) | RMSE (cm): 3.97 cm to 3.86 cm | Improved |
| add correlative fusion (CF) on top of prompts (ablation) | DuCos · RGB-D-D (real-world) | RMSE (cm): 0.1 cm | Improved |
| add gradient regulation (GR) on top of prompts (ablation) | DuCos · RGB-D-D (real-world) | RMSE (cm): 0.11 cm | Improved |
| combine CF and GR (ablation) | DuCos base model · RGB-D-D (real-world) | RMSE (cm): 0.15 cm | Improved |
| add Lagrangian duality (LD) strategy (ablation) | DuCos · RGB-D-D (real-world) | RMSE (cm): 3.68 cm | Improved |
| add alignment loss L_cf in CF (ablation) | DuCos · RGB-D-D (real-world) | RMSE (cm): 3.82 cm to 3.76 cm | Improved |
| add min-max normalization in GR (ablation) | DuCos · RGB-D-D (real-world) | RMSE (cm): 0.05 cm | Improved |
| more CF iterations (ablation) | DuCos CF module · RGB-D-D (real-world) | RMSE | Improved |
| use Pearson correlation fusion instead of addition (ablation) | CF module · RGB-D-D (real-world) | RMSE (cm): 0.02 cm | Improved |
| use Pearson correlation fusion instead of concatenation (ablation) | CF module · RGB-D-D (real-world) | RMSE (cm): 0.03 cm | Improved |
| use larger DA v2-L prompt instead of DA v2-S (ablation) | DuCos · NYU v2 (synthetic, x16) | RMSE: 7.37 -> 7.15 | Improved |
| use UniDepth v2-L prompt instead of UniDepth v2-S (ablation) | DuCos · NYU v2 (synthetic, x16) | RMSE: 8.69 -> 8.86 | Hurt |
Later papers that compared with this method
1 later accepted paper reports DuCos as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- DegBins: Degradation-Driven Binning for Depth Super-Resolution (NeurIPS 2026 · better on 15 of 15 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
- Dual-branch Graph Feature Learning for NLOS Imaging (AAAI 2025 · similarity 0.85)
- DegBins: Degradation-Driven Binning for Depth Super-Resolution (NeurIPS 2026 · similarity 0.84)
- PatchFusion: An End-to-End Tile-Based Framework for High-Resolution Monocular Metric Depth Estimation (CVPR 2024 · similarity 0.84)
- Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior (AAAI 2025 · similarity 0.84)
- C2PD: Continuity-Constrained Pixelwise Deformation for Guided Depth Super-Resolution (AAAI 2025 · similarity 0.84)
- Metric Depth Estimation from Arbitrarily Degraded Low-Resolution Depth Prompts (NeurIPS 2026 · similarity 0.84)
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.
