WAVE-Stereo: Warp-Aligned Volume Encoding for Stereo Matching

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. $\mathbf{\text{GeoWarp}}\mathbf{\text{Correspondence}}\mathbf{\text{Encoder}}\mathbf{\text{(GWCE)}}$ encodes matching search, residual alignment, and disparity prior in parallel at the ConvGRU input. To mitigate matching degradation in textureless regions, we propose $\mathbf{\text{Periodic}}\mathbf{\text{Global}}\mathbf{\text{Context}}\mathbf{\text{Propagation}}\mathbf{\text{(PGCP)}}$, 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.