MatchAttention: Matching the Relative Positions for High-Resolution Cross-View Matching
This work addresses the problem of efficient and accurate cross-view matching for applications like stereo vision, enabling real-time high-resolution processing with broad impact in computer vision.
The paper tackles the challenge of high-resolution cross-view matching by proposing MatchAttention, an attention mechanism that dynamically matches relative positions, and achieves state-of-the-art performance with low computational cost, such as ranking 1st on the Middlebury benchmark with 29ms inference time for KITTI-resolution.
Cross-view matching is fundamentally achieved through cross-attention mechanisms. However, matching of high-resolution images remains challenging due to the quadratic complexity and lack of explicit matching constraints in the existing cross-attention. This paper proposes an attention mechanism, MatchAttention, that dynamically matches relative positions. The relative position determines the attention sampling center of the key-value pairs given a query. Continuous and differentiable sliding-window attention sampling is achieved by the proposed BilinearSoftmax. The relative positions are iteratively updated through residual connections across layers by embedding them into the feature channels. Since the relative position is exactly the learning target for cross-view matching, an efficient hierarchical cross-view decoder, MatchDecoder, is designed with MatchAttention as its core component. To handle cross-view occlusions, gated cross-MatchAttention and a consistency-constrained loss are proposed. These two components collectively mitigate the impact of occlusions in both forward and backward passes, allowing the model to focus more on learning matching relationships. When applied to stereo matching, MatchStereo-B ranked 1st in average error on the public Middlebury benchmark and requires only 29ms for KITTI-resolution inference. MatchStereo-T can process 4K UHD images in 0.1 seconds using only 3GB of GPU memory. The proposed models also achieve state-of-the-art performance on KITTI 2012, KITTI 2015, ETH3D, and Spring flow datasets. The combination of high accuracy and low computational complexity makes real-time, high-resolution, and high-accuracy cross-view matching possible. Project page: https://github.com/TingmanYan/MatchAttention.