Research map / 3D vision / Depth and stereo
StableDepth: Scene-Consistent and Scale-Invariant Monocular Depth
Zheng Zhang, Lihe Yang, Tianyu Yang, Chaohui Yu, Xiaoyang Guo, Yixing Lao, Hengshuang Zhao
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 unlabeled pseudo-labeled video data with dual DPT decoders (ablation) | labeled-only DAv2-L baseline · Bonn | AbsRel: 0.071 -> 0.057 | Improved |
| add unlabeled pseudo-labeled video data with dual DPT decoders (ablation) | labeled-only DAv2-L baseline · ScanNet | AbsRel: 0.102 -> 0.097 | Improved |
| use single DPT decoder with reciprocal output for video data instead of dual decoders (ablation) | StableDepth · Bonn | AbsRel: 0.057 -> 0.109 | Hurt |
| use lightweight CNN head instead of second DPT decoder (ablation) | StableDepth · ScanNet | delta1: 0.913 -> 0.829 | Hurt |
| add second independent DPT decoder for unlabeled video (ablation) | single-DPT variant · Bonn | delta1: 0.914 -> 0.976 | Improved |
| add pseudo-labeled SA-V videos to Hypersim training (ablation) | Hypersim-only model · Sintel | AbsRel: 0.309 -> 0.326 | Hurt |
| add pseudo-labeled SA-V videos to Hypersim training (ablation) | Hypersim-only model · KITTI | AbsRel: 0.122 -> 0.143 | Hurt |
| add pseudo-labeled SA-V videos to Hypersim training (ablation) | Hypersim-only model · Bonn | AbsRel: 0.071 -> 0.057 | Improved |
| add VKITTI2, LightWheel, Hypersim-U and MatrixCity data on top of SA-V (ablation) | StableDepth · KITTI | AbsRel: 0.143 -> 0.114 | Improved |
| add video-trained dual-decoder StableDepth training to baseline for temporal consistency (ablation) | DAv2-L metric baseline · DAVIS | MTD: 0.003777 -> 0.003482 | Improved |
| use DAv2 instead of DINOv2 initialization (labeled data only) (ablation) | labeled-only model · Bonn | AbsRel: 0.086 -> 0.071 | Improved |
| use DAv2 instead of DINOv2 initialization (labeled+unlabeled) (ablation) | StableDepth · Bonn | AbsRel: 0.068 -> 0.057 | Improved |
Closest papers by meaning
- DCDepth: Progressive Monocular Depth Estimation in Discrete Cosine Domain (NeurIPS 2024 · similarity 0.90)
- UniDepth: Universal Monocular Metric Depth Estimation (CVPR 2024 · similarity 0.89)
- CH3Depth: Efficient and Flexible Depth Foundation Model with Flow Matching (CVPR 2025 · similarity 0.88)
- SharpDepth: Sharpening Metric Depth Predictions Using Diffusion Distillation (CVPR 2025 · similarity 0.88)
- MonSter: Marry Monodepth to Stereo Unleashes Power (CVPR 2025 · similarity 0.88)
- Video Depth Anything: Consistent Depth Estimation for Super-Long Videos (CVPR 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.
