Research map / 3D vision / Depth and stereo
DF-Net: Unsupervised Joint Learning of Depth and Flow using Cross-Task Consistency
Yuliang Zou, Zelun Luo, Jia-Bin Huang
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 forward-backward consistency loss (ablation) | DF-Net depth · KITTI raw Eigen test (trained on KITTI, monocular) | Abs Rel: 0.160 -> 0.150 | Improved |
| add cross-task consistency loss (ablation) | DF-Net depth · KITTI raw Eigen test (trained on KITTI, monocular) | Abs Rel: 0.160 -> 0.150 | Improved |
| add cross-task consistency loss (ablation) | DF-Net depth · KITTI raw Eigen test (trained on Cityscapes+KITTI, monocular) | Abs Rel: 0.155 -> 0.146 | Improved |
| add forward-backward consistency loss (ablation) | DF-Net depth · KITTI raw Eigen test (trained on Cityscapes+KITTI, monocular) | Abs Rel: 0.159 -> 0.146 | Improved |
| add cross-task consistency loss (ablation) | DF-Net depth · KITTI raw Eigen test (trained on Cityscapes+KITTI, monocular) | Sq Rel: 1.181 -> 1.182 | No clear effect |
| add cross-task consistency loss (ablation) | DF-Net flow · KITTI 2012 train | EPE: 4.70 -> 3.54 | Improved |
| add forward-backward consistency loss (ablation) | DF-Net flow · KITTI 2012 train | EPE: 3.86 -> 3.54 | Improved |
| add cross-task consistency loss (ablation) | DF-Net flow · KITTI 2015 train | F1: 28.37% -> 26.01% | Improved |
| add forward-backward consistency loss (ablation) | DF-Net flow · KITTI 2015 train | F1: 26.27% -> 26.01% | Improved |
| add cross-task consistency loss (ablation) | DF-Net flow · KITTI 2012 test | EPE: 5.8 -> 4.4 | Improved |
| add Cityscapes to KITTI training data | DF-Net depth · KITTI raw Eigen test | Abs Rel: 0.150 -> 0.146 | Improved |
| use monocular video only versus stereo supervision | DF-Net depth · KITTI raw Eigen test | Abs Rel: 0.150 -> 0.121 | Hurt |
Closest papers by meaning
- Feature-Level Collaboration: Joint Unsupervised Learning of Optical Flow, Stereo Depth and Camera Motion (CVPR 2021 · similarity 0.92)
- GeoNet: Unsupervised Learning of Dense Depth, Optical Flow and Camera Pose (CVPR 2018 · similarity 0.91)
- Learning Optical Flow From Still Images (CVPR 2021 · similarity 0.90)
- RM-Depth: Unsupervised Learning of Recurrent Monocular Depth in Dynamic Scenes (CVPR 2022 · similarity 0.90)
- Competitive Collaboration: Joint Unsupervised Learning of Depth, Camera Motion, Optical Flow and Motion Segmentation (CVPR 2019 · similarity 0.90)
- Flow-NeRF: Joint Learning of Geometry, Poses, and Dense Flow within Unified Neural Representations (CVPR 2025 · similarity 0.87)
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.
