Embedded Block Residual Network: A Recursive Restoration Model for Single-Image Super-Resolution

Qiu, Yajun; Wang, Ruxin; Tao, Dapeng; Cheng, Jun

doi:10.1109/iccv.2019.00428

Cited by 123 publications

(59 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Liu et al [29] proposed a kind of non-local module to capture deep feature correlations between each location and its neighborhood and employed the recurrent neural network structure for deep feature propagation. Qiu et al [30] proposed an embedded block residual network where different modules restore the information of different frequencies for a texture SR. Hu et al [31] proposed a channel-wise and spatial feature modulation network where LR features can be transformed to high informative features using feature-modulation memory modules. Jing et al [32] took the LR image and its downsampled resolution (DR) and upsampled resolution (UR) versions as inputs and learned the internal structure coherence with the pairs of UR-LR and LR-DR to generate a hierarchical dictionary.…”

Section: Related Workmentioning

confidence: 99%

Color-Guided Depth Map Super-Resolution Using a Dual-Branch Multi-Scale Residual Network with Channel Interaction

Chen

Gao

2020

Sensors

View full text Add to dashboard Cite

We designed an end-to-end dual-branch residual network architecture that inputs a low-resolution (LR) depth map and a corresponding high-resolution (HR) color image separately into the two branches, and outputs an HR depth map through a multi-scale, channel-wise feature extraction, interaction, and upsampling. Each branch of this network contains several residual levels at different scales, and each level comprises multiple residual groups composed of several residual blocks. A short-skip connection in every residual block and a long-skip connection in each residual group or level allow for low-frequency information to be bypassed while the main network focuses on learning high-frequency information. High-frequency information learned by each residual block in the color image branch is input into the corresponding residual block in the depth map branch, and this kind of channel-wise feature supplement and fusion can not only help the depth map branch to alleviate blur in details like edges, but also introduce some depth artifacts to feature maps. To avoid the above introduced artifacts, the channel interaction fuses the feature maps using weights referring to the channel attention mechanism. The parallel multi-scale network architecture with channel interaction for feature guidance is the main contribution of our work and experiments show that our proposed method had a better performance in terms of accuracy compared with other methods.

show abstract

Section: Related Workmentioning

confidence: 99%

Color-Guided Depth Map Super-Resolution Using a Dual-Branch Multi-Scale Residual Network with Channel Interaction

Chen

Gao

2020

Sensors

View full text Add to dashboard Cite

show abstract

“…With a different focus from these, Wang et al [53] analyse multiple Gaussian degradations in an attempt to reduce the reconstruction error in real-world data, Qin et al [54] combine the ideas of the channel, and spatial attention for building a deep multilevel residual attention network and lastly, Wu et al [55] come up with a novel perceptual loss for upsampling. Though most recent, however, none of these techniques were able to cross the benchmark established by an earlier but current state-of-the-art method [39].…”

Section: Prior Workmentioning

confidence: 99%

“…Similar to the methods [1], [22], [28], [39], [62], WDN is also trained on the DIV2K dataset by Timofte et al [65]. DIV2K dataset contains 1000 images at 2K resolution, among which 800 are for training, and 100 are for validation.…”

Section: Experiments and Analysismentioning

confidence: 99%

“…Performance on scaling factors of two and three: This computation requires the following modification in WDN: For a scale of two, we assume that the given input low-resolution image already has the dimension 2h×2w×1, and hence we (a) [10] (b) [12] (c) [14] (d) [22] (e) [24] (f) [25] (g) [39] (h) WDN (i) Ground truth (j) GT full (k) [10] (l) [12] (m) [14] (n) [22] (o) [24] (p) [25] (q) [39] (r) WDN (s) Ground truth (t) GT full Fig. 11: Predicted image visualisation for qualitative comparison on two images from the Urban100 dataset [70] for 4× super-resolution.…”

Section: Experiments and Analysismentioning

confidence: 99%

“…Quantitative comparison (4×) with state-of-the-art methods. Evaluation procedure used to compute the values is as used/described by Qiu et al[39]. All values have been captured from the non-self-ensemble variant.…”

mentioning

confidence: 99%

See 2 more Smart Citations

Going Much Wider with Deep Networks for Image Super-Resolution

Singh

Ramnath

Arunachalam

et al. 2020

2020 IEEE Winter Conference on Applications of Computer Vision (WACV)

View full text Add to dashboard Cite

Divide and conquer is an established algorithm design paradigm that has proven itself to solve a variety of problems efficiently. However, it is yet to be fully explored in solving problems with a neural network, particularly the problem of image super-resolution. In this work, we propose an approach to divide the problem of image super-resolution into multiple sub-problems and then solve/conquer them with the help of a neural network. Unlike a typical deep neural network, we design an alternate network architecture that is much wider (along with being deeper) than existing networks and is specially designed to implement the divide-and-conquer design paradigm with a neural network. Additionally, a technique to calibrate the intensities of feature map pixels is being introduced. Extensive experimentation on five datasets reveals that our approach towards the problem and the proposed architecture generate better and sharper results than current state-of-the-art methods.

show abstract

NAS-DIP: Learning Deep Image Prior with Neural Architecture Search

Chen

Gao

Robb

et al. 2020

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Super-ResDenoising Inpainting Dehazing Translation Fig. 1: Applications. We propose to learn deep image prior using a neural architecture search. The resulting network can be applied to solve various inverse image problems without training the model with a large-scale dataset with ground truth. Through extensive experimental evaluations, we show that our model compares favorably against existing hand-crafted CNN models for learning-free image restoration tasks and in some cases even reaches competitive performance when compared with recent learning-based models.

show abstract

Embedded Block Residual Network: A Recursive Restoration Model for Single-Image Super-Resolution

Cited by 123 publications

References 28 publications

Color-Guided Depth Map Super-Resolution Using a Dual-Branch Multi-Scale Residual Network with Channel Interaction

Color-Guided Depth Map Super-Resolution Using a Dual-Branch Multi-Scale Residual Network with Channel Interaction

Going Much Wider with Deep Networks for Image Super-Resolution

NAS-DIP: Learning Deep Image Prior with Neural Architecture Search

Contact Info

Product

Resources

About