Research map / 3D vision / Depth and stereo
M^2Depth: Self-supervised Two-Frame Multi-camera Metric Depth Estimation
Yingshuang Zou, Yikang Ding, Xi Qiu, Haoqian Wang, Haotian 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 spatial cost volume to mono prior (ablation) | M2Depth · DDAD | Abs. Rel.: 0.216 -> 0.212 | Improved |
| add temporal cost volume on top of spatial volume (ablation) | M2Depth · DDAD | Abs. Rel.: 0.212 -> 0.197 | Improved |
| add spatial-temporal fusion (STF) module (ablation) | M2Depth · DDAD | Abs. Rel.: 0.197 -> 0.194 | Improved |
| add depth edge loss (ablation) | M2Depth · DDAD | Abs. Rel.: 0.194 -> 0.192 | Improved |
| add multi-grained feature fusion (SAM features) in mono prior (ablation) | M2Depth · DDAD | Abs. Rel.: 0.192 -> 0.188 | Improved |
| use SAM features in both MFF and depth decoding (ablation) | M2Depth · DDAD | Abs. Rel.: 0.188 -> 0.183 | Improved |
| MFF attention fusion instead of simple addition (VFF) (ablation) | M2Depth · DDAD | Abs. Rel.: 0.185 -> 0.183 | Improved |
| SAM context features instead of FPN context features in decoder (ablation) | M2Depth · DDAD | Abs. Rel.: 0.196 -> 0.191 | Improved |
| predict front-camera pose instead of concatenating all views (ablation) | M2Depth · DDAD | Abs. Rel.: 0.189 -> 0.183 | Improved |
| adaptive depth sampling around mono prior vs fixed range (ablation) | M2Depth · DDAD | Abs. Rel.: 0.195 -> 0.183 | Improved |
| increase depth bins from 16 to 32 (ablation) | M2Depth · DDAD | Abs. Rel.: 0.194 -> 0.200 | No clear effect |
| increase input frames from two to three or four (ablation) | M2Depth · DDAD | Abs. Rel.: 0.183 -> 0.185 (3 frames), 0.186 (4 frames) | Hurt |
Later papers that compared with this method
3 later accepted papers report M2Depth as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- OmniNWM: Unifying the State-Action-Reward Triad for Closed-Loop Panoramic Driving Navigation World Models (ECCV 2026 · better on 2 of 2 shared results)
- Doe-2: 3D Representation World Model for Unified Driving Scene Forecasting (ECCV 2026 · better on 2 of 2 shared results)
- DiST-4D: Disentangled Spatiotemporal Diffusion with Metric Depth for 4D Driving Scene Generation (ICCV 2025 · better on 1 of 1 shared result)
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.
- R3D3* — R3D3: Dense 3D Reconstruction of Dynamic Scenes from Multiple Cameras (ICCV 2023) · 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
- The Temporal Opportunist: Self-Supervised Multi-Frame Monocular Depth (CVPR 2021 · similarity 0.90)
- ProDepth: Boosting Self-Supervised Multi-Frame Monocular Depth with Probabilistic Fusion (ECCV 2024 · similarity 0.88)
- SQLdepth: Generalizable Self-Supervised Fine-Structured Monocular Depth Estimation (AAAI 2024 · similarity 0.88)
- UniDepth: Universal Monocular Metric Depth Estimation (CVPR 2024 · similarity 0.87)
- STViT: Improving Self-Supervised Multi-Camera Depth Estimation with Spatial-Temporal Context and Adversarial Geometry Regularization (Student Abstract) (AAAI 2024 · similarity 0.86)
- SharpDepth: Sharpening Metric Depth Predictions Using Diffusion Distillation (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.
