MAFNet:Multi-frequency Adaptive Fusion Network for Real-time Stereo Matching
This addresses the deployment challenge of stereo matching for real-time applications on mobile devices, representing a strong domain-specific improvement.
The paper tackles the problem of real-time stereo matching on resource-constrained devices by proposing MAFNet, which uses efficient 2D convolutions and frequency-domain filtering to achieve high-quality disparity maps. The method significantly outperforms existing real-time approaches on Scene Flow and KITTI 2015 datasets.
Existing stereo matching networks typically rely on either cost-volume construction based on 3D convolutions or deformation methods based on iterative optimization. The former incurs significant computational overhead during cost aggregation, whereas the latter often lacks the ability to model non-local contextual information. These methods exhibit poor compatibility on resource-constrained mobile devices, limiting their deployment in real-time applications. To address this, we propose a Multi-frequency Adaptive Fusion Network (MAFNet), which can produce high-quality disparity maps using only efficient 2D convolutions. Specifically, we design an adaptive frequency-domain filtering attention module that decomposes the full cost volume into high-frequency and low-frequency volumes, performing frequency-aware feature aggregation separately. Subsequently, we introduce a Linformer-based low-rank attention mechanism to adaptively fuse high- and low-frequency information, yielding more robust disparity estimation. Extensive experiments demonstrate that the proposed MAFNet significantly outperforms existing real-time methods on public datasets such as Scene Flow and KITTI 2015, showing a favorable balance between accuracy and real-time performance.