Research map / 3D vision / Depth and stereo
Self-Supervised Relative Depth Learning for Urban Scene Understanding
Huaizu Jiang, Gustav Larsson, Michael Maire Greg Shakhnarovich, Erik Learned-Miller
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 relative depth self-supervised pre-training | AlexNet FCN32s · CityScapes segmentation | mean IoU: 40.7 -> 45.4 | Improved |
| add relative depth self-supervised pre-training | AlexNet FCN32s · CamVid segmentation | mean IoU: 44.0 -> 53.4 | Improved |
| add relative depth pre-training on CityDriving | VGG16 FCN8s · CityScapes segmentation | mean IoU: 51.6 -> 59.8 | Improved |
| add relative depth pre-training on CityDriving | VGG16 FCN8s · KITTI segmentation | mean IoU: 34.2 -> 48.6 | Improved |
| replace ImageNet supervised pre-training with relative depth pre-training | VGG16 FCN8s · CityScapes segmentation | mean IoU: 63.4 -> 59.8 | Hurt |
| add unsupervised domain adaptation by fine-tuning depth task on CityScapes videos (ablation) | Ours CD · CityScapes FCN32s | mean IoU: 55.0 -> 56.2 | Improved |
| add unsupervised domain adaptation by fine-tuning depth task on KITTI videos (ablation) | Ours CD · KITTI FCN32s | mean IoU: 45.6 -> 46.0 | Improved |
| add unsupervised domain adaptation on KITTI videos (ablation) | Ours CD · CamVid FCN8s | mean IoU: 60.3 -> 60.1 | No clear effect |
| add domain adaptation on KITTI videos (ablation) | Ours CD · road segmentation F1 | F1: 94.74 -> 95.66 | Improved |
| add domain adaptation on KITTI videos (ablation) | Ours CD · car detection easy | AP: 92.84 -> 94.31 | Improved |
| replace ImageNet pre-training with relative depth pre-training plus KITTI adaptation | VGG16 U-Net · KITTI monocular depth | RMSE: 5.546 -> 5.317 | Improved |
| replace ImageNet pre-training with relative depth pre-training plus KITTI adaptation | ResNet50 U-Net · KITTI monocular depth | RMSE: 5.073 -> 4.903 | Improved |
Closest papers by meaning
- Self-Supervised Monocular Depth Estimation: Solving the Dynamic Object Problem by Semantic Guidance (ECCV 2020 · similarity 0.90)
- Mining Supervision for Dynamic Regions in Self-Supervised Monocular Depth Estimation (CVPR 2024 · similarity 0.89)
- The Temporal Opportunist: Self-Supervised Multi-Frame Monocular Depth (CVPR 2021 · similarity 0.89)
- From-Ground-To-Objects: Coarse-to-Fine Self-supervised Monocular Depth Estimation of Dynamic Objects with Ground Contact Prior (CVPR 2024 · similarity 0.88)
- Improved Monocular Depth Prediction Using Distance Transform Over Pre-semantic Contours with Self-supervised Neural Networks (CVPR 2025 · similarity 0.88)
- Unsupervised Semantic Segmentation Through Depth-Guided Feature Correlation and Sampling (CVPR 2024 · 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.
