AI & ML Breaks Assumption

Proves that stereo matching can reach state-of-the-art performance without the computationally heavy cost volumes used by almost all modern methods.

March 27, 2026

Original Paper

WAFT-Stereo: Warping-Alone Field Transforms for Stereo Matching

Yihan Wang, Jia Deng

arXiv · 2603.24836

The Takeaway

WAFT-Stereo replaces cost volumes with simple warping, resulting in models that are up to 6.7x faster while ranking first on major benchmarks. This challenges the foundational architectural assumption that explicit cost-volume construction is necessary for accurate depth perception.

From the abstract

We introduce WAFT-Stereo, a simple and effective warping-based method for stereo matching. WAFT-Stereo demonstrates that cost volumes, a common design used in many leading methods, are not necessary for strong performance and can be replaced by warping with improved efficiency. WAFT-Stereo ranks first on ETH3D, KITTI and Middlebury public benchmarks, reducing the zero-shot error by 81% on ETH3D benchmark, while being 1.8-6.7x faster than competitive methods. Code and model weights are available