Reproduce “GA-Net: Guided Aggregation Net for End-to-end Stereo Matching” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Feihu Zhang, Victor Prisacariu, Ruigang Yang, Philip H.S. Torr
Researchers at the University of Oxford and Baidu Research developed GA-Net, an end-to-end stereo matching network that integrates novel differentiable semi-global (SGA) and local guided aggregation (LGA) layers. This architecture achieves state-of-the-art accuracy on datasets like Scene Flow and KITTI by efficiently refining disparity estimation, particularly in difficult regions, while significantly reducing computational costs compared to networks relying on 3D convolutions.
Key insights
- Traditional, non-differentiable stereo aggregation techniques can be effectively adapted into learnable, context-adaptive layers within end-to-end deep learning frameworks.
- Guided aggregation layers offer significantly lower computational complexity (O(KN) vs. O(K^3CN) for 3D convolutions), enabling more efficient and faster inference in stereo networks.
- Dynamically generated aggregation weights, based on local image content, enable more precise disparity refinement and enhance accuracy in challenging regions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.