Research map / Architectures and efficiency / RNNs and normalization
Practical Deep Stereo (PDS): Toward applications-friendly deep stereo matching
Stepan Tulyakov, Anton Ivanov, François Fleuret
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 |
|---|---|---|---|
| train on full-size 960 × 540 images instead of 512 × 256 patches (ablation) | PDS · FlyingThings3D, test size 960 × 540 | 3PE (%): 5.28 -> 4.50 | Improved |
| train on full-size 960 × 540 images instead of 512 × 256 patches (ablation) | PDS · FlyingThings3D, test size 960 × 540 | MAE (px): 3.55 -> 3.40 | Improved |
| replace SoftArgmin with Sub-pixel MAP at inference (ablation) | PDS trained with L1 loss · FlyingThings3D, disparity range [0, 255] | 3PE (%): 4.50 -> 4.22 | Improved |
| replace SoftArgmin with Sub-pixel MAP at inference (ablation) | PDS trained with L1 loss · FlyingThings3D, disparity range [0, 255] | MAE (px): 3.40 -> 3.42 | Hurt |
| extend the disparity range from [0, 255] to [0, 511] without retraining (ablation) | PDS with SoftArgmin · FlyingThings3D | 3PE (%): 4.50 -> 5.20 | Hurt |
| extend the disparity range from [0, 255] to [0, 511] without retraining (ablation) | PDS with Sub-pixel MAP · FlyingThings3D | 3PE (%): 4.22 -> 4.27 | No clear effect |
| train with sub-pixel cross-entropy instead of L1 loss (ablation) | PDS with Sub-pixel MAP · FlyingThings3D, disparity range [0, 255] | 3PE (%): 4.22 -> 3.80 | Improved |
| train with sub-pixel cross-entropy instead of L1 loss (ablation) | PDS with Sub-pixel MAP · FlyingThings3D, disparity range [0, 255] | MAE (px): 3.42 -> 3.63 | Hurt |
Closest papers by meaning
- Stereo Matching Using Multi-Level Cost Volume and Multi-Scale Feature Constancy (TPAMI 2019 · similarity 0.86)
- Efficient Deep Learning for Stereo Matching (CVPR 2016 · similarity 0.86)
- Open-World Stereo Video Matching with Deep RNN (ECCV 2018 · similarity 0.85)
- A Survey on Deep Learning Techniques for Stereo-Based Depth Estimation (TPAMI 2020 · similarity 0.85)
- Stereo Matching by Training a Convolutional Neural Network to Compare Image Patches (JMLR 2016 · similarity 0.84)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 2024 · similarity 0.84)
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.
