Research map / Label-efficient and robust learning / Self-supervised and contrastive
Learning to Discriminate Scene Structures Makes Self-Supervised Depth Learning Scalable
Mengtan Zhang, Yuchuan Cui, Yu Ma, Zizhan Guo, Wei Ye, Rui Fan
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes MaSS, a self-supervised monocular depth framework that learns to discriminate scene structures so that depth learning scales across diverse training scenes.
Problem
Self-supervised monocular depth estimation often degrades when trained on diverse scenes, because models encode dataset-specific structural biases that cause cross-domain optimization conflicts.
Approach
Under self-supervision, models encode dataset-specific structural biases rather than transferable scene geometry; MaSS is designed to address this by learning to discriminate scene structures so that training scales across diverse data. Full method details are not available from the abstract.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add scene-structure contrastive learning objective (ablation) | MaSS · Waymo zero-shot | Abs Rel: 0.163 -> 0.123 | Improved |
| add decoupled structural pathway (representation decoupling) (ablation) | MaSS · NYUv2 (mixed training) | Abs Rel: 0.116 -> 0.104 | Improved |
| add bidirectional interaction architecture between depth encoder and VFM (ablation) | MaSS · Waymo zero-shot | Abs Rel: 0.190 -> 0.123 | Improved |
| use MaSS instead of frozen DINOv2 with DPT head (ablation) | DINOv2 backbone · nuScenes (mixed training) | Abs Rel: 0.186 -> 0.134 | Improved |
| fully fine-tune DINOv2 with DPT instead of freezing it (ablation) | DINOv2 + DPT · 7-Scenes zero-shot | delta1: 0.839 -> 0.600 | Hurt |
| replace DINOv2 with stronger DINOv3 backbone (ablation) | MaSS · Training sets and zero-shot | Abs Rel: 0.134 -> 0.136 (nuScenes) | No clear effect |
| train on mixed datasets instead of single dataset | MaSS · KITTI, DDAD, NYUv2, nuScenes | Abs Rel: 0.094 -> 0.092 (KITTI) | Improved |
| train on mixed datasets instead of single dataset | existing SSMDE baselines (Lite-Mono, Dynamo-Depth, SC-DepthV3, DCPI-Depth) · NYUv2 | Abs Rel: 0.151 -> 0.331 (DCPI-Depth) | Hurt |
| mixed-dataset training instead of KITTI-only | MaSS · Waymo zero-shot | Abs Rel: 0.141 -> 0.123 | Improved |
| add 50K TartanAir samples (150K to 200K) | MaSS mixed · DIML and DIODE out-of-distribution | Abs Rel, delta1: +4.3% Abs Rel, +3.0% delta1 | Improved |
| add 50K TartanAir samples (150K to 200K) | MaSS mixed · training domains | Abs Rel, delta1: +1.5% Abs Rel, +0.2% delta1 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- Depth Anything V2 — Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data (CVPR 2024) · better on 1 of 2 shared results
- Jasmine — Jasmine: Harnessing Diffusion Prior for Self-supervised Depth Estimation (NeurIPS 2025) · 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.
Surprising findings
- Under self-supervision, models tend to encode dataset-specific structural biases rather than transferable scene geometry, causing cross-domain optimization conflicts.
Closest papers by meaning
- Digging Into Self-Supervised Monocular Depth Estimation (ICCV 2019 · similarity 0.91)
- The Temporal Opportunist: Self-Supervised Multi-Frame Monocular Depth (CVPR 2021 · similarity 0.91)
- Excavating the Potential Capacity of Self-Supervised Monocular Depth Estimation (ICCV 2021 · similarity 0.91)
- Fine-Grained Semantics-Aware Representation Enhancement for Self-Supervised Monocular Depth Estimation (ICCV 2021 · similarity 0.91)
- SQLdepth: Generalizable Self-Supervised Fine-Structured Monocular Depth Estimation (AAAI 2024 · similarity 0.90)
- Mining Supervision for Dynamic Regions in Self-Supervised Monocular Depth Estimation (CVPR 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.
