Image Restoration Using Very Deep Convolutional Encoder-Decoder Networks with Symmetric Skip Connections

By Xiao-Jiao Mao, Chunhua Shen, Yubin Yang. 2016

1,698 citations recorded by the source corpus.

How researchers see this paper

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

Citations in the corpus

Memory-Efficient Hierarchical Neural Architecture Search for Image Denoising (2019, methodology): “Motivated by this, most recent works on image denoising have shifted their approaches to deep learning, which builds a mapping function from noisy images to the desired corresponding clean images with deep learning models and have often outperformed conventional methods signiﬁcantly [27, 33, 22].”
Audio Codec Enhancement with Generative Adversarial Networks (2020, background): “GANs have achieved tremendous success in the field of computer vision to generate high-resolution diverse images from complex datasets [18], image restoration [19], as well as for speech enhancement [20].”
Enhanced Deep Residual Networks for Single Image Super-Resolution (2017, background): “In [16], they argue that those nested skip connections provide fast and improved convergence.”
Neural Nearest Neighbors Networks (2018, methodology): “Hence, for many dense prediction tasks, e. g. in image analysis and restoration, stacking ever more convolutional blocks has remained the prevailing choice to obtain bigger receptive ﬁelds [20, 22, 31, 39, 51].”
Imaging through glass diffusers using densely connected convolutional networks (2017, methodology): “This is the “encoder–decoder network” architecture [42,43], where the dense and downsampling transition blocks serve as encoder to extract the feature maps from the input patterns, and the dense and upsampling transition blocks serve as decoder to perform pixel-wise regression.”
Deep Learning for Environmentally Robust Speech Recognition (2017, methodology): “For image restoration and further image processing tasks, deep convolutional encoder-decoder networks were proposed in Mao et al. (2016) and delivered promising performance.”
Content-aware image restoration: pushing the limits of fluorescence microscopy (2018, methodology): “1We use networks of moderate size (≈ 10(6) parameters) based on the U-net architecture [36, 37], together with a per-pixel similarity loss (e.”
Scale-Recurrent Network for Deep Image Deblurring (2018, methodology): “Encoder-decoder Network Encoder-decoder network [24, 27] refers to those symmetric CNN structures that ﬁrst progressively transform input data into feature maps with smaller spatial size and more channels (encoder), and then transform them back to the shape of the input (decoder).”
Residual Dense Network for Image Restoration (2018, methodology): “Methods LapSRN DRRN MemNet MDSR EDSR RDN [13] [24] [25] [16] [16] (ours) # param.”
Dynamic Scene Deblurring With Parameter Selective Sharing and Nested Skip Connections (2019, methodology): “After the seminal work of U-net [25], skip connections between the corresponding encoder and decoder stages were widely used as an effective architecture for pixel-wise regression in optical flow estimation [3], image restoration [18] and raindrop removal [24].”
Human-Aware Motion Deblurring (2019, methodology): “Numerous CNN-based methods have also been applied in an endto-end fashion for image processing and generation problems, such as segmentation [24, 42], super-resolution [53, 22, 14, 15, 44], denoising [25, 49, 59, 47, 48], dehazing and deraining [32, 55], enhancement [31, 56] etc.”
Deep Convolutional Framelets: A General Deep Learning Framework for Inverse Problems (2017, background): “ose artifacts from reduced tube currents, the streaking artifacts originated from sparse projection views show globalized artifacts that are difﬁcult to remove using conventional denoising CNNs [12], [13]. Jin et al [14] and Han et al [15] independently proposed a residual learning using U-Net [2] t”
Toward Convolutional Blind Denoising of Real Photographs (2018, methodology): “Benefited from the modeling capability of CNNs, the studies [61, 38, 55] show that it is feasible to learn a single model for blind Gaussian denoising.”
Low-Light Image Enhancement with Wavelet-Based Diffusion Models (2023, methodology): “Following the development of learning-based image restoration methods [Dong et al. 2015; Mao et al. 2016; Sun et al. 2015], LLNet [Lore et al.”
Cooperative Semantic Segmentation and Image Restoration in Adverse Environmental Conditions (2019, background): “[28] propose a 30-layer convolutional auto-encoder network named RED by introducing symmetric skip connections for image denoising”
Deep Learning for Image Super-Resolution: A Survey (2019, methodology): “Since the transposed convolution enlarges the image size in an end-to-end manner while maintaining a connectivity pattern compatible with vanilla convolution, it is widely used as upsampling layers in SR models [57], [78], [79], [85].”
Neighbor2Neighbor: Self-Supervised Denoising from Single Noisy Images (2021, background): “Numerous CNNbased image denoisers, e.g., DnCNN [34], U-Net [26], RED [20], MemNet [28], and SGN [10], have superior performance over traditional denoisers.”