Lightweight Image Super-Resolution with Information Multi-distillation Network

By Zheng Hui, Xinbo Gao, Yunchu Yang, Xiumei Wang. 2019

1,190 citations recorded by the source corpus.

In recent years, single image super-resolution (SISR) methods using deep convolution neural network (CNN) have achieved impressive results. Thanks to the powerful representation capabilities of the deep networks, numerous previous ways can learn the complex non-linear mapping between low-resolution (LR) image patches and their high-resolution (HR) versions. However, excessive convolutions will limit the application of super-resolution technology in low computing power devices. Besides, super-resolution of any arbitrary scale factor is a critical issue in practical applications, which has not been well solved in the previous approaches. To address these issues, we propose a lightweight information multi-distillation network (IMDN) by constructing the cascaded information multi-distillation blocks (IMDB), which contains distillation and selective fusion parts. Specifically, the distillation module extracts hierarchical features step-by-step, and fusion module aggregates them according to the importance of candidate features, which is evaluated by the proposed contrast-aware channel attention mechanism. To process real images with any sizes, we develop an adaptive cropping strategy (ACS) to super-resolve block-wise image patches using the same well-trained model. Extensive experiments suggest that the proposed method performs favorably against the state-of-the-art SR algorithms in term of visual quality, memory footprint, and inference time. Code is available at \urlhttps://github.com/Zheng222/IMDN.

Subfields: Computer Vision and Pattern Recognition, Media Technology, Efficiency & Compression

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Luc Van Gool
2. Radu Timofte
3. Lei Zhang
4. Chao Dong
5. Kai Zhang
6. Chen Change Loy
7. Ming-Hsuan Yang
8. Yu Qiao

In Media Technology

1. Radu Timofte
2. Chao Dong
3. Luc Van Gool
4. Kai Zhang
5. Lei Zhang
6. Yulun Zhang
7. Jinjin Gu
8. Jingyun Liang

In Efficiency & Compression

1. Lei Zhang
2. Yulun Zhang
3. Jinjin Gu
4. Yun Raymond Fu
5. Kai Zhang
6. Juncheng Li
7. Yuan Xie
8. Yu Qiao

Citations in the corpus

• Designing a Practical Degradation Model for Deep Blind Image Super-Resolution (2021, background): “With the advance of deep neural networks (DNNs), there is a dramatic upsurge of using feed-forward DNNs for fast and effective SISR [16, 22, 24, 45, 56].”
• SwinIR: Image Restoration Using Swin Transformer (2021, methodology): “Method Scale #Params #Mult-Adds Set5 [3] Set14 [87] BSD100 [58] Urban100 [34] Manga109 [60]PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM PSNR SSIM CARN [2] ×2 1,592K 222.8G 37.76 0.9590 33.52 0.9166 32.09 0.8978 31.92 0.9256 38.36 0.9765 FALSR-A [12] ×2 1,021K 234.7G 37.82 0.959 33.55 0.9168 32.1 0.8987 3”
• Efficient Transformer for Single Image Super-Resolution (2021, methodology): “IMDN [11] improves the architecture of IDN and introduces the information multidistillation blocks to extract the hierarchical features effectively.”
• AIM 2019 Challenge on Constrained Super-Resolution: Methods and Results (2019, neutral): “The proposed IMDN, as illustrated in Figure 1, is built upon their recently published work [8].”
• Interpreting Super-Resolution Networks with Local Attribution Maps (2020, methodology): “For networks from the literature, we select: SRCNN [11], FSRCNN, EDSR, LapSRN [23], SRResNet [24], DDBPN [18], RDN [51], MSRN [25], RCAN, RRDBNet, CARN, SAN, IMDN [20], RNAN [50], RFDN [28], PAN [52], DRLN, and CSNLN [31].”
• Efficient Image Super-Resolution Using Pixel Attention (2020, methodology): “IMDN [12] extracts hierarchical features step-by-step by split operations, and then aggregates them by simply using a 1 × 1 convolution.”
• LatticeNet: Towards Lightweight Image Super-Resolution with Lattice Block (2020, methodology): “Inspired by the success of SENet [14] and IMDN [16], we utilize two statistics to compute the combination coefficients: the mean and the standard deviation of a feature map.”
• Residual Feature Distillation Network for Lightweight Image Super-Resolution (2020, methodology): “All the values are calculated on the Y channel of the YCbCr channels converted from the RGB channels as with existing works [12,22,17,34,32,1,11,10].”
• Efficient Long-Range Attention Network for Image Super-resolution (2022, methodology): “We first compare our ELAN-light with state-of-the-art light-weight SR models, including CNN-based models CARN[2], IMDN [18], LAPAR-A[25], Lat-ticeNet[37], and transformer-based models ESRT [36] and SwinIR-light[28].”
• Transformer for Single Image Super-Resolution (2021, methodology): “In addition, ESRT achieves better performance Scale IMDN [16] LK-KPN [6] ESRT (Ours) PSNR SSIM PSNR SSIM PSNR SSIM”