Reproduce “Iterative Filter Adaptive Network for Single Image Defocus Deblurring” 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
Junyong Lee, Hyeongseok Son, Jaesung Rim, Sunghyun Cho, Seungyong Lee
POSTECH researchers developed an Iterative Filter Adaptive Network (IFAN) for single image defocus deblurring, which efficiently handles spatially-varying and large blur by iteratively applying learned separable 1D filters. The method achieves state-of-the-art deblurring performance through a training strategy incorporating defocus disparity estimation and a reblurring task, outperforming prior approaches with fewer computational resources.
Key insights
- Iterative application of small 1D separable filters (IAC) efficiently expands the effective receptive field, allowing the network to manage large defocus blur with reduced computational cost compared to dense 2D filters.
- Training the network to estimate dual-pixel disparity maps helps it learn more accurate per-pixel blur magnitudes, which improves the restoration of image textures.
- Enforcing a reblurring consistency constraint on the predicted filters ensures they accurately capture blur characteristics, thereby enhancing the overall fidelity of the deblurred images.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.