Research map / Label-efficient and robust learning / Recognition and long-tail
CFNet: Cascade and Fused Cost Volume for Robust Stereo Matching
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Use fused low-resolution cost volumes rather than a single volume (ablation) | CFNet · KITTI 2015, Middlebury, ETH3D validation | D1_all / bad 2.0 / bad 1.0: 1.79 -> 1.71; 22.65 -> 22.27; 3.67 -> 3.57 | Helped |
| Use pyramid feature extraction instead of a Resnet-like network (ablation) | CFNet · KITTI 2015, Middlebury, ETH3D validation | D1_all / bad 2.0 / bad 1.0: 1.76 -> 1.71; 22.81 -> 22.27; 3.49 -> 3.57 | Mixed |
| Replace uniform sampling with uncertainty-based adaptive disparity sampling (ablation) | CFNet · KITTI 2015, Middlebury, ETH3D validation | D1_all / bad 2.0 / bad 1.0: 1.92 -> 1.71; 23.8 -> 22.27; 3.97 -> 3.57 | Helped |
| Learn uncertainty parameters rather than setting them as hyperparameters (ablation) | CFNet · KITTI 2015, Middlebury, ETH3D validation | D1_all / bad 2.0 / bad 1.0: 1.78 -> 1.71; 23.13 -> 22.27; 3.83 -> 3.57 | Helped |
| Set all stage loss weights to 1 rather than use Loss1:1, Loss2:1, Loss3:0.5 (ablation) | CFNet · KITTI 2015, Middlebury, ETH3D validation | D1_all / bad 2.0 / bad 1.0: 1.71 -> 1.84; 22.27 -> 23.87; 3.57 -> 3.47 | Mixed |
| Use three-stage finetuning instead of two-stage finetuning (ablation) | CFNet · KITTI 2015, Middlebury, ETH3D validation | D1_all / bad 2.0 / bad 1.0: 1.70 -> 1.71; 22.77 -> 22.27; 3.99 -> 3.57 | Mixed |
| Augment small datasets during the third finetuning stage (ablation) | CFNet · KITTI 2015, Middlebury, ETH3D validation | D1_all / bad 2.0 / bad 1.0: 1.70 -> 1.71; 22.57 -> 22.27; 3.92 -> 3.57 | Mixed |
| Train CFNet on Scene Flow and evaluate on real datasets without adaptation | CFNet · KITTI 2012, KITTI 2015, Middlebury, ETH3D cross-domain | D1_all / bad 2.0 / bad 1.0: 4.7 vs 6.2; 5.8 vs 6.5; 28.2 vs 21.8; 5.8 vs 6.2 | Mixed |
| Remove the 1% most uncertain pixels using the uncertainty estimate (ablation) | CFNet · KITTI 2015 validation | D1_all error rate: from 1.55% to 1.09% | Helped |
Closest papers by meaning
- AANet: Adaptive Aggregation Network for Efficient Stereo Matching (CVPR 2020 · similarity 0.91)
- Cascade Cost Volume for High-Resolution Multi-View Stereo and Stereo Matching (CVPR 2020 · similarity 0.91)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 2024 · similarity 0.90)
- Any-Stereo: Arbitrary Scale Disparity Estimation for Iterative Stereo Matching (AAAI 2024 · similarity 0.89)
- Modeling Stereo-Confidence out of the End-to-End Stereo-Matching Network via Disparity Plane Sweep (AAAI 2024 · similarity 0.89)
- IINet: Implicit Intra-inter Information Fusion for Real-Time Stereo Matching (AAAI 2024 · similarity 0.89)
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.