Fast and Accurate Single Image Super-Resolution via Information Distillation Network

By Zheng Hui, Xiumei Wang, Xinbo Gao. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018

876 citations recorded by the source corpus.

Recently, deep convolutional neural networks (CNNs) have been demonstrated remarkable progress on single image super-resolution. However, as the depth and width of the networks increase, CNN-based super-resolution methods have been faced with the challenges of computational complexity and memory consumption in practice. In order to solve the above questions, we propose a deep but compact convolutional network to directly reconstruct the high resolution image from the original low resolution image. In general, the proposed model consists of three parts, which are feature extraction block, stacked information distillation blocks and reconstruction block respectively. By combining an enhancement unit with a compression unit into a distillation block, the local long and short-path features can be effectively extracted. Specifically, the proposed enhancement unit mixes together two different types of features and the compression unit distills more useful information for the sequential blocks. In addition, the proposed network has the advantage of fast execution due to the comparatively few numbers of filters per layer and the use of group convolution. Experimental results demonstrate that the proposed method is superior to the state-of-the-art methods, especially in terms of time performance. Code is available at https://github.com/Zheng222/IDN-Caffe.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/8f684080d2b81d3178d681d6917cb077c082a9e1

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

In Media Technology

Citations in the corpus

Densely Residual Laplacian Super-Resolution (2019, background): “To obtain distinct features at multiple scales, multi-branch networks [25], [26], [27] are proposed.”
Channel-Wise and Spatial Feature Modulation Network for Single Image Super-Resolution (2018, methodology): “However, many very deep networks, such as VDSR [15], LapSRN [27], EDSR [12] and IDN [31], adopt single-path direct connections or short skip connections among layers, where hierarchical features could hardly be fully utilized and long-term information that provides some clues for SR would be lost…”
Feature Normalized Knowledge Distillation for Image Classification (2020, methodology): “Recently, KD has made huge success, some works have extended this effective idea to other application domains [13, 39, 18], others [37, 17, 24] improve the standard KD by kinds of techniques such as feature based distillation and achieve better results.”
MADNet: A Fast and Lightweight Network for Single-Image Super Resolution (2020, methodology): “Note that we reimplement IDN [15] with PyTorch, and the official TensorFlow implementation is at https://github.”
Learning Enriched Features for Real Image Restoration and Enhancement (2020, background): “In contrast to directly producing a latent HR image, recent SR networks [56,95,94,48]”
Feedback Network for Image Super-Resolution (2019, background): “In other words, the tasks of our network are to suppress the smooth area of the original input image[16] and to predict high-frequency components (i.”
Deep Learning for Single Image Super-Resolution: A Brief Review (2018, background): “Recently, some works on designing light models [105], [109], [110] and learning sparse structural NN [111] were proposed to achieve relatively good performance with less storage and computation, which are very meaningful in practice.”
Deep Learning for Image Super-Resolution: A Survey (2019, methodology): “reduces the number of parameters and operations at the expense of a little performance loss [28], [98].”
Efficient Transformer for Single Image Super-Resolution (2021, methodology): “IDN [12] compresses the model size by using the group convolution and combine short-term and long-term features.”
LatticeNet: Towards Lightweight Image Super-Resolution with Lattice Block (2020, methodology): “In this section, we conduct extensive experiments on four publicly available SR benchmark datasets and compare with 12 state-of-the-art lightweight SR models: SRCNN [8], FSRCNN [9], VDSR [18], DRCN [19], LapSRN [20], DRRN [30], MemNet [31], IDN* [17] 4 , CARN [2], BSRN [5], FALSR [6], and IMDN [16].”
Transformer for Single Image Super-Resolution (2021, methodology): “IDN [17] compresses the model size by using the group convolution and combining short-term and long-term features.”
Dynamic Scene Deblurring With Parameter Selective Sharing and Nested Skip Connections (2019, background): “Input images are often added to the reconstructed ones in image/video restoration [10, 30, 32, 9], since learning the residual image through a CNN is much easier than reconstructing decent output.”
A Deep Journey into Super-resolution (2019, methodology): “The Information Distillation Network (IDN) [69] consists of three blocks: a feature extraction block, multiple stacked information distillation blocks and a reconstruction block.”
Learning Enriched Features for Fast Image Restoration and Enhancement (2022, background): “In contrast to directly producing a latent HR image, recent SR networks [59], [60], [61], [62] employ the residual learning framework [63] to learn the high-frequency image detail, which is later added to the input LR image to produce the ﬁnal result.”
Simultaneous Enhancement and Super-Resolution of Underwater Imagery for Improved Visual Perception (2020, background): “Additionally, inspired by the success of deep residual networks for terrestrial SISR [73, 42, 30], several models have been proposed for underwater SISR in recent years [12, 34], which report exciting results with reasonable computational overhead.”
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].”
Dual Reconstruction Nets for Image Super-Resolution with Gradient Sensitive Loss (2018, methodology): “For example, in [18] they first train the network with `MAE and then fine-tune it by `MSE.”
Lightweight Image Super-Resolution with Information Multi-distillation Network (2019, methodology): “[11] proposed an information distillation network (IDN) that explicitly divided the preceding extracted features into two parts, one was retained and another was further processed.”