Research map / Label-efficient and robust learning / Recognition and long-tail
GraftNet: Towards Domain Generalized Stereo Matching With a Broad-Spectrum and Task-Oriented Feature
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 |
|---|---|---|---|
| Replace concatenation cost volume with cosine-similarity cost volume (ablation) | PSMNet trained on SceneFlow · KITTI 2015 | EPE: 3.24 -> 2.98 | Helped |
| Add normalized pure-similarity cost rather than concatenated features (ablation) | PSMNet · KITTI 2015 | >3px error rate: 19.5% -> 15.4% | Helped |
| Graft VGG's ImageNet feature (ablation) | PSMNet with cosine-similarity cost volume · KITTI 2015 | EPE: 2.98 -> 1.86 | Helped |
| Add a feature adaptor to the grafted VGG feature (ablation) | Graft-PSMNet · KITTI 2015 | EPE: 1.86 -> 1.47 | Helped |
| Retrain the cost aggregation module using the adapted feature (ablation) | Graft-PSMNet · KITTI 2015 | EPE: 1.47 -> 1.32 | Helped |
| Graft VGG's ImageNet feature without the adaptor (ablation) | GANet-11 with cosine-similarity cost volume · Middlebury | EPE: 6.09 -> 7.75 | Hurt |
| Add a feature adaptor to the grafted VGG feature (ablation) | Graft-GANet · Middlebury | EPE: 7.75 -> 1.96 | Helped |
| Use color-jitter augmentation during training | Graft-PSMNet · ETH3D | >1px error rate: 10.7% -> 7.7% | Helped |
| Use cosine similarity instead of L2 distance for the cost volume (ablation) | PSMNet · KITTI 2015 | EPE: 2.86 -> 2.98 | Hurt |
| Use a U-Net adaptor instead of a linear adaptor (ablation) | Graft-PSMNet · KITTI 2015 | >3px error rate: 6.06% -> 5.60% | Helped |
| Use a ResNet50 DCL feature instead of VGG16 classification feature (ablation) | Graft-PSMNet · KITTI 2015 | >3px error rate: 6.39% -> 9.17% | Hurt |
| Use a broad-spectrum VGG feature without task adaptation (ablation) | GANet-11 · ETH3D | >1px error rate: 7.08% -> 13.2% | Hurt |
Closest papers by meaning
- Domain Generalized Stereo Matching via Hierarchical Visual Transformation (CVPR 2023 · similarity 0.92)
- Revisiting Domain Generalized Stereo Matching Networks From a Feature Consistency Perspective (CVPR 2022 · similarity 0.91)
- Masked Representation Learning for Domain Generalized Stereo Matching (CVPR 2023 · similarity 0.90)
- Hierarchical Object-Aware Dual-Level Contrastive Learning for Domain Generalized Stereo Matching (NeurIPS 2024 · similarity 0.90)
- Robust Synthetic-to-Real Transfer for Stereo Matching (CVPR 2024 · similarity 0.89)
- Stereo Risk: A Continuous Modeling Approach to Stereo Matching (ICML 2024 · 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.