Multi-Temporal Recurrent Neural Networks For Progressive Non-Uniform Single Image Deblurring With Incremental Temporal Training

TL;DR

This paper introduces a multi-temporal recurrent neural network approach for non-uniform single image deblurring, utilizing incremental temporal training to improve results over traditional multi-scale methods, especially for severe blurs.

Contribution

The paper proposes a novel multi-temporal recurrent neural network with incremental training for progressive non-uniform image deblurring, outperforming existing multi-scale methods.

Findings

Outperforms state-of-the-art multi-scale methods on GoPro dataset in PSNR.

Uses fewer parameters than existing methods.

Demonstrates effective progressive deblurring over iterations.

Abstract

Multi-scale (MS) approaches have been widely investigated for blind single image / video deblurring that sequentially recovers deblurred images in low spatial scale first and then in high spatial scale later with the output of lower scales. MS approaches have been effective especially for severe blurs induced by large motions in high spatial scale since those can be seen as small blurs in low spatial scale. In this work, we investigate alternative approach to MS, called multi-temporal (MT) approach, for non-uniform single image deblurring. We propose incremental temporal training with constructed MT level dataset from time-resolved dataset, develop novel MT-RNNs with recurrent feature maps, and investigate progressive single image deblurring over iterations. Our proposed MT methods outperform state-of-the-art MS methods on the GoPro dataset in PSNR with the smallest number of parameters.