Residual Dense Network for Image Super-Resolution

By Yulun Zhang, Yapeng Tian, Yu Kong, Bineng Zhong, Yun Raymond Fu. 2018

3,919 citations recorded by the source corpus.

A very deep convolutional neural network (CNN) has recently achieved great success for image super-resolution (SR) and offered hierarchical features as well. However, most deep CNN based SR models do not make full use of the hierarchical features from the original low-resolution (LR) images, thereby achieving relatively-low performance. In this paper, we propose a novel residual dense network (RDN) to address this problem in image SR. We fully exploit the hierarchical features from all the convolutional layers. Specifically, we propose residual dense block (RDB) to extract abundant local features via dense connected convolutional layers. RDB further allows direct connections from the state of preceding RDB to all the layers of current RDB, leading to a contiguous memory (CM) mechanism. Local feature fusion in RDB is then used to adaptively learn more effective features from preceding and current local features and stabilizes the training of wider network. After fully obtaining dense local features, we use global feature fusion to jointly and adaptively learn global hierarchical features in a holistic way. Experiments on benchmark datasets with different degradation models show that our RDN achieves favorable performance against state-of-the-art methods.

Subfields: Computer Vision and Pattern Recognition, Media Technology

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

Designing a Practical Degradation Model for Deep Blind Image Super-Resolution (2021, background): “Although signiﬁcant improvements on the PSNR [27, 61] and perceptual quality [24, 49] have been achieved, such methods usually do not perform well on real images.”
Fast, Accurate and Lightweight Super-Resolution with Neural Architecture Search (2019, background): “Therefore, too deep and large models such as RDN [17], RCAN [18] are excluded here.”
ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks (2018, methodology): “ierent levels. A similar network structure is proposed in [36] that also applies a multilevel residual network. However, our RRDB diers from [36] in that we use dense block [34] in the main path as [11], where the network capacity becomes higher beneting from the dense connections. In addition to th”
Image Super-Resolution Using Very Deep Residual Channel Attention Networks (2018, methodology): “, EDSR [10] and RDN [17]), we pay more attention to our proposed RIR, CA, and the basic module RCAB.”
Pre-Trained Image Processing Transformer (2020, background): “Urban100 (4×):img 012 HR Bicubic VDSR [41] EDSR [51] RDN [94]”
Correction to: Single Image Super-Resolution via a Holistic Attention Network (2020, methodology): “Following the protocols of [38,37,41], we further com-”
A Survey of Deep Learning for Scientific Discovery (2020, background): “More recently, Residual Dense Networks [270] have been a popular approach for super-resolution on standard benchmarks (with code available https://github . com/yulunzhang/RDN ), as well as Predictive Filter Flow [114], (code: https://github . com/aimerykong/predictive-filter-flow ) which has also…”
Image denoising using deep CNN with batch renormalization (2020, background): “In general, different network architectures can extract different features [67].”
GridDehazeNet: Attention-Based Multi-Scale Network for Image Dehazing (2019, methodology): “The grid network has clear advantages over the encoderdecoder network and the conventional multi-scale network extensively used in image restoration [18, 41, 38, 27].”
Unsupervised Degradation Representation Learning for Blind Super-Resolution (2021, background): “Zhang et al. [45] then combined residual learning and dense connection to construct a residual dense network (RDN) with over 100 layers.”
Pyramid Attention Network for Semantic Segmentation (2018, positive): “Encoder-decoder: State-of-the-art segmentation frameworks are mostly based on encoderdecoder networks[25][1][31][24], which have also been successfully applied to many computer vision tasks, including human pose estimation[22], object detection[17][18], image stylization[28], portrait matting[27][26”
Meta-SR: A Magnification-Arbitrary Network for Super-Resolution (2019, methodology): “Actually, our Meta-Upscale Module can be incorporated into most previous methods [35, 34, 17] by replacing the typical upscale module.”
Wide Activation for Efficient and Accurate Image Super-Resolution (2018, methodology): “It is also introduced in SISR networks in SRResNet [17].”
Efficient Image Super-Resolution Using Pixel Attention (2020, methodology): “Recently, many deep neural networks [45,44,3,20,43,9] have been introduced to improve the reconstruction results.”
Multi-Scale Progressive Fusion Network for Single Image Deraining (2020, background): “The signiﬁcance of these features produced at different stages has been veriﬁed on image reconstruction tasks [28, 43].”
Avoiding Overfitting: A Survey on Regularization Methods for Convolutional Neural Networks (2022, background): “One relevant work in this area is the Residual Dense Network, which has a version for super-resolution [74] and image denoising [75] purposes.”
Multi-Scale Boosted Dehazing Network With Dense Feature Fusion (2020, methodology): “Numerous image restoration methods fuse features by using dense connection [61], feature concatenation [64] or weighted element-wise summation [9, 62].”
Progressive Fusion Video Super-Resolution Network via Exploiting Non-Local Spatio-Temporal Correlations (2019, methodology): “During training, a data agumentation scheme (random flip and rotation) from [21, 41] is adopted.”
Structure-Preserving Super Resolution With Gradient Guidance (2020, methodology): “[52] propose RDN by utilizing residual dense blocks in the SR framework.”