Research map / 3D vision / Depth and stereo
Unsupervised Adaptation for Deep Stereo
Alessio Tonioni, Matteo Poggi, Stefano Mattoccia, Luigi Di Stefano
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 unsupervised adaptation with AD-CENSUS targets and confidence-guided loss | DispNet · KITTI 2015 train | bad 3 (%): 7.46 -> 4.02 | Improved |
| add unsupervised adaptation with SGM targets and confidence-guided loss | DispNet · KITTI 2015 train | bad 3 (%): 7.46 -> 4.21 | Improved |
| add unsupervised adaptation with SGM targets | DispNet · Middlebury 2014 (quarter res) | bad 1 (%): 32.82 -> 22.91 | Improved |
| add unsupervised adaptation with AD-CENSUS targets | DispNet · Middlebury 2014 (quarter res) | bad 1 (%): 32.82 -> 25.38 | Improved |
| add supervised fine-tuning on KITTI 2012 ground truth | DispNet · Middlebury 2014 (quarter res) | bad 1 (%): 32.82 -> 40.21 | Hurt |
| add supervised fine-tuning on KITTI 2012 ground truth | DispNet · KITTI 2015 train | bad 3 (%): 7.46 -> 4.58 | Improved |
| replace supervised KITTI 2012 fine-tuning with unsupervised adaptation (CENSUS) on the same frames | DispNet · KITTI 2015 train (generalization) | bad 3 (%): 4.58 -> 4.34 | Improved |
| replace supervised KITTI 2012 fine-tuning with unsupervised adaptation (CENSUS) | DispNet · KITTI 2012 train | bad 3 (%): 2.89 -> 4.29 | Hurt |
| raise confidence threshold tau up to 0.99 (ablation) | DispNet unsupervised adaptation · KITTI 2012 train / KITTI 2015 (10000 steps) | bad 3 (%) | Improved |
| add smoothness term with lambda = 0.1 (ablation) | DispNet unsupervised adaptation · KITTI 2015 generalization | bad 3 (%) | Improved |
| increase smoothness weight lambda beyond 0.6 (ablation) | DispNet unsupervised adaptation · KITTI 2012 train | bad 3 (%) | Hurt |
| use only stereo pairs with ground truth (000000_10) instead of all KITTI 2015 pairs (ablation) | DispNet CENSUS adaptation · KITTI 2015 train | bad 3 (%): 0.09% error rate and 0.04% avg disparity error increase | No clear effect |
Closest papers by meaning
- Learning to Adapt for Stereo (CVPR 2019 · similarity 0.93)
- Continual Adaptation for Deep Stereo (TPAMI 2021 · similarity 0.92)
- Real-Time Self-Adaptive Deep Stereo (CVPR 2019 · similarity 0.91)
- Learning Stereo from Single Images (ECCV 2020 · similarity 0.91)
- Unsupervised Domain Adaptation for Depth Prediction from Images (TPAMI 2019 · similarity 0.91)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 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.
