LoFormer: Local Frequency Transformer for Image Deblurring

TL;DR

LoFormer is a novel transformer architecture that models long-range dependencies in image deblurring by combining local frequency domain self-attention with an MLP gating mechanism, improving detail preservation and global context understanding.

Contribution

Introduces LoFormer, a local frequency domain self-attention transformer with an MLP gating mechanism for enhanced image deblurring performance.

Findings

Achieves 34.09 dB PSNR on GoPro dataset

Significantly improves deblurring quality over existing methods

Efficiently balances coarse and fine-grained feature learning

Abstract

Due to the computational complexity of self-attention (SA), prevalent techniques for image deblurring often resort to either adopting localized SA or employing coarse-grained global SA methods, both of which exhibit drawbacks such as compromising global modeling or lacking fine-grained correlation. In order to address this issue by effectively modeling long-range dependencies without sacrificing fine-grained details, we introduce a novel approach termed Local Frequency Transformer (LoFormer). Within each unit of LoFormer, we incorporate a Local Channel-wise SA in the frequency domain (Freq-LC) to simultaneously capture cross-covariance within low- and high-frequency local windows. These operations offer the advantage of (1) ensuring equitable learning opportunities for both coarse-grained structures and fine-grained details, and (2) exploring a broader range of representational properties compared to coarse-grained global SA methods. Additionally, we introduce an MLP Gating mechanism complementary to Freq-LC, which serves to filter out irrelevant features while enhancing global learning capabilities. Our experiments demonstrate that LoFormer significantly improves performance in the image deblurring task, achieving a PSNR of 34.09 dB on the GoPro dataset with 126G FLOPs. https://github.com/DeepMed-Lab-ECNU/Single-Image-Deblur