KBNet: Kernel Basis Network for Image Restoration

TL;DR

KBNet introduces a novel kernel basis attention module that adaptively aggregates spatial information using learnable kernels, leading to state-of-the-art image restoration performance with reduced computational costs.

Contribution

The paper proposes the KBNet architecture with a kernel basis attention module and multi-axis feature fusion, offering adaptive spatial aggregation and improved efficiency for image restoration.

Findings

Achieves state-of-the-art results on multiple image restoration benchmarks.

Requires less computational cost than previous SOTA methods.

Effectively models local structures with learnable kernel bases.

Abstract

How to aggregate spatial information plays an essential role in learning-based image restoration. Most existing CNN-based networks adopt static convolutional kernels to encode spatial information, which cannot aggregate spatial information adaptively. Recent transformer-based architectures achieve adaptive spatial aggregation. But they lack desirable inductive biases of convolutions and require heavy computational costs. In this paper, we propose a kernel basis attention (KBA) module, which introduces learnable kernel bases to model representative image patterns for spatial information aggregation. Different kernel bases are trained to model different local structures. At each spatial location, they are linearly and adaptively fused by predicted pixel-wise coefficients to obtain aggregation weights. Based on the KBA module, we further design a multi-axis feature fusion (MFF) block to encode and fuse channel-wise, spatial-invariant, and pixel-adaptive features for image restoration. Our model, named kernel basis network (KBNet), achieves state-of-the-art performances on more than ten benchmarks over image denoising, deraining, and deblurring tasks while requiring less computational cost than previous SOTA methods.