NFResNet: Multi-scale and U-shaped Networks for Deblurring

TL;DR

This paper introduces NFResNet and NFResNet+ architectures for image deblurring, incorporating a novel NFResblock with Fourier and non-linear activation layers, demonstrating significant improvements in PSNR and SSIM on a video deblurring dataset.

Contribution

The paper proposes NFResblock and two new architectures, NFResNet and NFResNet+, for enhanced multi-scale and U-shaped image deblurring, with extensive experimental validation.

Findings

Significant PSNR and SSIM improvements on the Deep Video Deblurring dataset.

Effective use of multiple loss functions for training.

Ablation studies validating each component's contribution.

Abstract

Multi-Scale and U-shaped Networks are widely used in various image restoration problems, including deblurring. Keeping in mind the wide range of applications, we present a comparison of these architectures and their effects on image deblurring. We also introduce a new block called as NFResblock. It consists of a Fast Fourier Transformation layer and a series of modified Non-Linear Activation Free Blocks. Based on these architectures and additions, we introduce NFResnet and NFResnet+, which are modified multi-scale and U-Net architectures, respectively. We also use three different loss functions to train these architectures: Charbonnier Loss, Edge Loss, and Frequency Reconstruction Loss. Extensive experiments on the Deep Video Deblurring dataset, along with ablation studies for each component, have been presented in this paper. The proposed architectures achieve a considerable increase in Peak Signal to Noise (PSNR) ratio and Structural Similarity Index (SSIM) value.