Research map / 3D vision / Depth and stereo
Reusable Architecture Growth for Continual Stereo Matching
Chenghao Zhang, Gaofeng Meng, Bin Fan, Kun Tian, Zhaoxiang Zhang, Shiming Xiang, Chunhong Pan
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 task-specific cell search instead of fixed cells (ablation) | RAG · DrivingStereo (supervised) | FAE EPE: 0.746 -> 0.600 | Improved |
| grow both Feature Net and Matching Net (ablation) | RAG · DrivingStereo (supervised) | FAE EPE: 0.735 -> 0.637 | Improved |
| grow Matching Net only vs Feature Net only (ablation) | RAG · DrivingStereo (supervised) | FAE EPE: 0.735 -> 0.672 | Improved |
| use non-zero initial iteration record c0=10 instead of 0 (ablation) | RAG · DrivingStereo (supervised) | EPE: 0.818 -> 0.655 | Improved |
| raise probability multiple gamma from 1 to 5 (ablation) | RAG · DrivingStereo (supervised) | EPE: 0.655 -> 0.703 | Hurt |
| use proposed validation score instead of error-only (ablation) | RAG · DrivingStereo (supervised) | EPE: 0.671 -> 0.660 | Improved |
| set target parameters to Phi/2 vs Phi/3 (ablation) | RAG · DrivingStereo (supervised) | EPE: 0.664 -> 0.637 | Improved |
| add scene contrastive loss to Scene Router (ablation) | Scene Router · DrivingStereo | scene division accuracy: about 15% | Improved |
| use Driving subset as proxy supervision instead of none (ablation) | RAG · DrivingStereo (self-supervised) | FAE EPE: 1.895 -> 1.709 | Improved |
| use Monkaa or Flyingthings3D as proxy supervision (ablation) | RAG · DrivingStereo (self-supervised) | FAE D1-all: 8.61% -> 9.97% | Hurt |
| use gamma=2 vs gamma=1 in self-supervised growth (ablation) | RAG · DrivingStereo (self-supervised) | EPE: 1.911 -> 1.834 | Improved |
| use larger target parameters phi=Phi in self-supervised (ablation) | RAG · DrivingStereo (self-supervised) | EPE: 1.813 -> 1.794 | Improved |
Closest papers by meaning
- Continual Adaptation for Deep Stereo (TPAMI 2021 · similarity 0.92)
- AANet: Adaptive Aggregation Network for Efficient Stereo Matching (CVPR 2020 · similarity 0.90)
- Hierarchical Neural Architecture Search for Deep Stereo Matching (NeurIPS 2020 · similarity 0.90)
- The Temporal Opportunist: Self-Supervised Multi-Frame Monocular Depth (CVPR 2021 · similarity 0.90)
- PPMStereo: Pick-and-Play Memory Construction for Consistent Dynamic Stereo Matching (NeurIPS 2025 · similarity 0.88)
- 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.
