| Full name | MatchAttention (Zero-shot) |
| Description | MatchAttention is a novel attention mechanism that embeds explicit matching constraints by using continuous relative positions to dynamically target the exact sampling center for key-value pairs. By leveraging Continuous Attention Sampling, it achieves differentiable, linear-complexity matching, enabling highly efficient and accurate high-resolution stereo inference. |
| Parameters | channels [384, 768, 1024, 1536], blocks [8, 8, 8, 2] |
| Publication title | MatchAttention: Embedding explicit matching constraints into attention for efficient stereo matching |
| Publication authors | Tingman Yan, Tao Liu, Chenghao Li, Xilian Yang, Qunfei Zhao, and Zeyang Xia |
| Publication venue | Arxiv, 2025 |
| Publication URL | https://arxiv.org/abs/2510.14260 |
| Programming language(s) | Pytorch, CUDA |
| Hardware | RTX 5090 |
| Source code or download URL | https://github.com/TingmanYan/MatchAttention |
| Submission creation date | 15 Aug, 2025 |
| Last edited | 4 Jun, 2026 |
