KBNet: Kernel Basis Network for Image Restoration

6 Mar 2023  ·  Yi Zhang, Dasong Li, Xiaoyu Shi, Dailan He, Kangning Song, Xiaogang Wang, Hongwei Qin, Hongsheng Li ·

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.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Grayscale Image Denoising BSD68 sigma15 KBNet PSNR 31.98 # 2
Compare
Grayscale Image Denoising BSD68 sigma25 KBNet PSNR 29.54 # 1
Compare
Grayscale Image Denoising BSD68 sigma50 KBNet PSNR 26.65 # 2
Compare
Color Image Denoising CBSD68 sigma15 KBNet PSNR 34.41 # 2
Compare
Color Image Denoising CBSD68 sigma25 KBNet PSNR 31.80 # 1
Compare
Color Image Denoising Kodak24 sigma50 KBNet PSNR 30.04 # 1
Compare
Color Image Denoising McMaster sigma50 KBNet PSNR 30.27 # 1
Compare
Grayscale Image Denoising Set12 sigma15 KBNet PSNR 33.40 # 1
Compare
Grayscale Image Denoising Set12 sigma25 KBNet PSNR 31.08 # 1
Compare
Grayscale Image Denoising Set12 sigma50 KBNet PSNR 28.04 # 1
Compare
Image Denoising SIDD KBNet PSNR (sRGB) 40.35 # 2
Compare
SSIM (sRGB) 0.972 # 2
Compare
Single Image Deraining Test1200 KBNet PSNR 33.82 # 2
Compare
SSIM 0.931 # 2
Compare
Single Image Deraining Test2800 KBNet PSNR 34.19 # 1
Compare
SSIM 0.944 # 1
Compare
Color Image Denoising urban100 sigma15 KBNet Average PSNR 35.15 # 1
Compare
Grayscale Image Denoising urban100 sigma15 KBNet PSNR 33.77 # 2
Compare
Grayscale Image Denoising Urban100 sigma25 KBNet PSNR 31.45 # 3
Compare
Color Image Denoising Urban100 sigma25 KBNet PSNR 32.96 # 1
Compare
Color Image Denoising Urban100 sigma50 KBNet PSNR 30.04 # 2
Compare
Grayscale Image Denoising Urban100 sigma50 KBNet PSNR 28.33 # 2
Compare

Methods
Add Remove

No methods listed for this paper. Add relevant methods here
Contact us on: hello@paperswithcode.com . Papers With Code is a free resource with all data licensed under CC-BY-SA.
Terms Data policy Cookies policy