Research map / 3D vision / Depth and stereo
Learning Monocular Depth by Distilling Cross-domain Stereo Networks
Xiaoyang Guo, Hongsheng Li, Shuai Yi, Jimmy Ren, Xiaogang Wang
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 |
|---|---|---|---|
| distill from a synthetic-pretrained stereo network instead of training monocular net directly on synthetic data | monocular depth network · KITTI Eigen (80m) | Abs Rel: 0.662 -> 0.109 | Improved |
| pretrain monocular network on Scene Flow before KITTI supervision | Direct Sup. monocular · KITTI Eigen (all of K) | Abs Rel: 0.105 -> 0.106 | No clear effect |
| pretrain monocular network on Scene Flow with 100 KITTI labels | Direct Sup. monocular · KITTI Eigen (100 of K) | Abs Rel: 0.187 -> 0.194 | Hurt |
| add unsupervised fine-tuning of the stereo proxy (ours) | StereoNoFt->Mono · KITTI Eigen (80m) | Abs Rel: 0.109 -> 0.105 | Improved |
| add ImageNet pretraining of monocular encoder | StereoUnsupFt->Mono · KITTI Eigen (80m) | Abs Rel: 0.105 -> 0.099 | Improved |
| add Cityscapes pretraining stage for monocular network | StereoUnsupFt->Mono pt · KITTI Eigen (80m) | Abs Rel: 0.099 -> 0.095 | Improved |
| use our occlusion-aware unsupervised loss instead of Godard et al. loss for stereo fine-tuning | StereoUnsupFt · KITTI Eigen stereo | Abs Rel: 0.076 -> 0.061 | Improved |
| use our occlusion-aware unsupervised loss instead of Godard et al. loss | StereoUnsupFt->Mono · KITTI Eigen distilled mono | Abs Rel: 0.116 -> 0.105 | Improved |
| use Godard et al. unsupervised loss to finetune stereo net instead of no fine-tuning | StereoNoFt · KITTI Eigen stereo | Abs Rel: 0.072 -> 0.076 | Hurt |
| add supervised fine-tuning with 100 images to the stereo net | StereoNoFt · KITTI Eigen stereo | RMS: 3.836 -> 3.325 | Improved |
| add unsupervised fine-tuning to the stereo net | StereoNoFt · Scene Flow test | MAE (px): 3.41 -> 4.23 | Hurt |
| add supervised 100-image fine-tuning to stereo before distillation | StereoNoFt->Mono · KITTI Eigen distilled mono | RMS: 4.656 -> 4.449 | Improved |
Closest papers by meaning
- Single View Stereo Matching (CVPR 2018 · similarity 0.91)
- Learning Monocular Depth Estimation Infusing Traditional Stereo Knowledge (CVPR 2019 · similarity 0.90)
- Unsupervised Monocular Depth Estimation With Left-Right Consistency (CVPR 2017 · similarity 0.90)
- CroMo: Cross-Modal Learning for Monocular Depth Estimation (CVPR 2022 · similarity 0.90)
- DepthFM: Fast Generative Monocular Depth Estimation with Flow Matching (AAAI 2025 · similarity 0.89)
- Distilling Monocular Foundation Model for Fine-grained Depth Completion (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.
