It is an excellent paper of Shenzhen and Taicang in 2022.
We propose a asymmetric super-resolution network for blind image super-solution.
Contributions: 1. An asymmtric architecture is used to enhance the effect of local key features on square kernel in horizontal and vertical ways for extracting more accurate low-frequency features and accelerating training speed in SISR.
2. Using a model can better deal with image super-resolution of certain scale, blind super-resolution and blind super-resolution with blind noise.
Abstract
Deep convolutional neural networks (CNNs) have been widely applied for low-level vision over the past five years. According to nature of different applications, designing appropriate CNN architectures is developed. However, customized architectures gather different features via treating all pixel points as equal to improve the performance of given application, which ignores the effects of local power pixel points and results in low training efficiency. In this paper, we propose an asymmetric CNN (ACNet) comprising an asymmetric block (AB), a memory enhancement block (MEB) and a high-frequency feature enhancement block (HFFEB) for image super-resolution. The AB utilizes one-dimensional asymmetric convolutions to intensify the square convolution kernels in horizontal and vertical directions for promoting the influences of local salient features for SISR. The MEB fuses all hierarchical low-frequency features from the AB via residual learning (RL) technique to resolve the long-term dependency problem and transforms obtained low-frequency features into high-frequency features. The HFFEB exploits low- and high-frequency features to obtain more robust super-resolution features and address excessive feature enhancement problem. Additionally, it also takes charge of reconstructing a high-resolution (HR) image. Extensive experiments show that our ACNet can effectively address single image super-resolution (SISR), blind SISR and blind SISR of blind noise problems. The code of the ACNet is shown at https://github.com/hellloxiaotian/ACNet.
Training a ACNet for image super-resolution with certain scale
x2
python x2/train.py --patch_size 64 --batch_size 16 --max_steps 660000 --decay 400000 --model acnet --ckpt_name acnet_x2 --ckpt_dir checkpoint/acnet_x2 --scale 2 --num_gpu 2 (If the number of gpu is 2, num_gpu may be set to 2, otherwise, it may be 1)
x3
python x3/train.py --patch_size 64 --batch_size 16 --max_steps 660000 --decay 400000 --model acnet --ckpt_name acnet_x3 --ckpt_dir checkpoint/acnet_x3 --scale 3 --num_gpu 2 (If the number of gpu is 2, num_gpu may be set to 2, otherwise, it may be 1)
x4
python x4/train.py --patch_size 64 --batch_size 16 --max_steps 660000 --decay 400000 --model acnet --ckpt_name acnet_x4 --ckpt_dir checkpoint/acnet_x4 --scale 4 --num_gpu 2 (If the number of gpu is 2, num_gpu may be set to 2, otherwise, it may be 1)
Training a ACNet for blind super-resolution as well as image super-resolution with varying scale
python acnet-b/train.py --patch_size 64 --batch_size 8 --max_steps 660000 --decay 400000 --model acnet-b --ckpt_name acnet_b --ckpt_dir checkpoint/acnet_b --scale 0 --num_gpu 2 (If the number of gpu is 2, num_gpu may be set to 2, otherwise, it may be 1)
