Search papers, labs, and topics across Lattice.
This paper introduces WAVE-Stereo, a novel stereo matching framework that integrates correlation volumes and warped features through a GeoWarp Correspondence Encoder (GWCE) to enhance matching accuracy. By implementing Periodic Global Context Propagation (PGCP), the method effectively addresses the challenge of matching degradation in textureless regions, leading to improved performance on various benchmarks. The results demonstrate that WAVE-Stereo achieves competitive zero-shot generalization accuracy and real-time inference speeds, highlighting its practical applicability in stereo matching tasks.
Achieving a remarkable 3.18% D1-all accuracy on KITTI 2015, WAVE-Stereo redefines stereo matching by seamlessly integrating correlation and warping cues for enhanced performance.
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.
