Reproduce “F2former: When Fractional Fourier Meets Deep Wiener Deconvolution and Selective Frequency Transformer for Image 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
Subhajit Paul, Sahil Kumawat, Ashutosh Gupta, Deepak Mishra
F2former introduces a deep learning architecture that integrates the Fractional Fourier Transform (FRFT) for image deblurring, leveraging FRFT's ability to handle spatially varying and non-uniform blur more effectively. The model establishes new state-of-the-art performance across various benchmarks, achieving up to 2 dB PSNR gains in defocus deblurring and consistent improvements in motion deblurring datasets.
Key insights
- The Fractional Fourier Transform (FRFT) provides a more flexible and unified spatial-frequency representation than the standard FT, proving superior for analyzing non-stationary signals and effectively capturing spatially varying artifacts in images.
- Integrating FRFT into both deconvolution (F2WD) and self-attention mechanisms (F2SA) within a deep learning framework leads to richer feature representations and significantly improved robustness against complex, non-uniform blur.
- Learnable fractional orders for FRFT and adaptive frequency-division multiplexing within the network allow for dynamic optimization of spatial-frequency representations and feature refinement tailored to specific blur characteristics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.