Research map / 3D vision / Depth and stereo
MetricAnything: Scaling Metric Depth Pretraining with Noisy Heterogeneous Sources
Jiahui Yang, Donglin Di, Xuancheng Zhang, Lei Fan, Jianxun Cui, Hao Li, Baorui Ma
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 |
|---|---|---|---|
| scale pretraining data from 5% to 100% (ablation) | MetricAnything-Pretrain · KITTI 8x depth super-resolution | AbsRel: 5.22 -> 2.34 | Improved |
| replace U-Net skip with inverse skip, pseudo labels (ablation) | Student model · NuScenes | Abs: 0.187 -> 0.125 | Improved |
| replace U-Net skip with inverse skip, pseudo labels (ablation) | Student model · ETH3D | Abs: 0.269 -> 0.182 | Improved |
| replace U-Net skip with inverse skip, real-world labels (ablation) | Student model · NuScenes | Abs: 0.213 -> 0.235 | Hurt |
| train with pseudo labels instead of real-world labels (ablation) | Student model · NuScenes | Abs: 0.235 -> 0.125 | Improved |
| use distance-balanced inverse-depth loss instead of inverse-depth (ablation) | MetricAnything · DIODE >50m | AbsRel: 0.632 -> 0.589 | Improved |
| use distance-balanced loss instead of inverse-depth at close range (ablation) | MetricAnything · DIODE 0-10m | AbsRel: 0.467 -> 0.465 | No clear effect |
| increase sparse prompt points from 500 to 64000 (ablation) | MetricAnything · Hypersim | AbsRel: 0.043 -> 0.031 | Improved |
| increase prompt points from 500 to 64000 (ablation) | MetricAnything · Hypersim | time (ms): 224 -> 308 | Hurt |
| increase balance coefficient C from 50 to 600 (ablation) | distance-balanced loss · DIODE 0-40m | AbsRel: 0.441 -> 0.465 | Hurt |
| increase balance coefficient C from 50 to 600 (ablation) | distance-balanced loss · DIODE >40m | AbsRel: 0.635 -> 0.559 | Improved |
| finetune pretrained model instead of training from scratch (ablation) | MetricAnything · nuScenes radar 0-50m | MAE (mm): 1335.4 -> 651.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 4 on most shared results.
- MoGe-2 — MoGe-2: Accurate Monocular Geometry with Metric Scale and Sharp Details (NeurIPS 2025) · better on 2 of 3 shared results
- Depth Pro — Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (ICLR 2025) · better on 1 of 2 shared results
- TacoDepth — TacoDepth: Towards Efficient Radar-Camera Depth Estimation with One-stage Fusion (CVPR 2025) · better on 1 of 1 shared result
- DUSt3R — DUSt3R: Geometric 3D Vision Made Easy (CVPR 2024) · 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
- SharpDepth: Sharpening Metric Depth Predictions Using Diffusion Distillation (CVPR 2025 · similarity 0.91)
- UniDepth: Universal Monocular Metric Depth Estimation (CVPR 2024 · similarity 0.90)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.89)
- Depth Anything V2 (NeurIPS 2024 · similarity 0.89)
- Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (ICLR 2025 · similarity 0.89)
- RSA: Resolving Scale Ambiguities in Monocular Depth Estimators through Language Descriptions (NeurIPS 2024 · similarity 0.89)
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.
