提出新立体匹配方法，融合相关图与特征扭曲优势，提升精度与速度。
WAVE-Stereo: Warp-Aligned Volume Encoding for Stereo Matching
- 并行编码相关图、残差对齐和视差先验，增强匹配搜索能力
- 在KITTI 2015上达3.18% D1-all，Booster上4.42% Bad-2.0，推理仅66ms
- 适合追求实时高精度立体匹配的视觉系统开发者
现有迭代立体匹配方法主要采用两种对应关系表示：通过相关体积显式搜索匹配，以及利用扭曲特征进行局部残差修正，但两者通常独立建模。本文提出WAVE-Stereo，核心思想是相关体积与特征扭曲提供互补匹配线索。提出GeoWarp对应编码器（GWCE），在ConvGRU输入层并行编码匹配搜索、残差对齐与视差先验。为缓解无纹理区域匹配退化问题，设计周期性全局上下文传播（PGCP），以周期方式传播全局空间信息。在五个真实世界基准测试——Middlebury、ETH3D、KITTI 2012、KITTI 2015和Booster上，WAVE-Stereo实现竞争力的零样本泛化性能，无需外部基础模型先验，达到KITTI 2015上3.18% D1-all、Booster上4.42% Bad-2.0，且推理时间仅66ms，兼顾精度与效率。代码已开源。
原文摘要 · Abstract (English)
Existing iterative stereo matching methods primarily adopt two types of correspondence representation: explicit matching search via correlation volumes and local residual refinement via warped features, yet the two remain separately modeled. We propose WAVE-Stereo, built on a core insight: correlation volumes and feature warping provide complementary matching cues. \textbf{GeoWarp Correspondence Encoder (GWCE)} encodes matching search, residual alignment, and disparity prior in parallel at the ConvGRU input. To mitigate matching degradation in textureless regions, we propose \textbf{Periodic Global Context Propagation (PGCP)}, which propagates global spatial information in a periodic manner. On five real-world benchmarks -- Middlebury, ETH3D, KITTI 2012, KITTI 2015, and Booster -- WAVE-Stereo achieves competitive zero-shot generalization accuracy without any external foundation model prior, achieving 3.18\% D1-all on KITTI 2015, 4.42\% Bad-2.0 on Booster, and 66ms real-time inference, striking a favorable balance between accuracy and efficiency. Our code is available at https://github.com/yamanoko-do/WAVE-Stereo.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。