Dual-Path Attention Network for Compressed Sensing Image Reconstruction

Sun, Yubao; Chen, Jiwei; Liu, Qingshan; Liu, Bo; Guo, Guodong

doi:10.1109/tip.2020.3023629

Cited by 104 publications

(49 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our proposed ISTA-Net ++ is compared with several representative state-of-the-art methods including BM3D-AMP [18], LDAMP [19], DIP [20], ReconNet [28], DPDNN [14], GDN [15], ISTA-Net + [2], NLR-CSNet [21], DPA-Net [10], and MAC-Net [9]. Different from other deep network-based end-to-end methods, for ISTA-Net ++ , we use the five sampling matrices for five CS ratios to train our model only once.…”

Section: Comparison With State-of-the-art Methodsmentioning

confidence: 99%

“…where k denotes the iteration index, and ρ is the step size. Fueled by the rise of deep learning, data-driven neural networks with diverse modules [7,8,9,10] have been proposed for image CS reconstruction by directly learning the inverse mapping from the CS measurement domain to the original signal domain. Most recently, some deep unfolding networks [2,11,12,13,14,15,16] are developed to combine the merits of both the model-and data-driven methods and yield a better signal recovery performance.…”

Section: % 50%mentioning

confidence: 99%

“…On the other hand, we find that most CS networks suffer from unbalanced reconstruction results in various scenes. The reason is that their training dataset Train91 [8] is unbalanced in the quantity distribution of different scenes, thus leading to undesirable results when reconstructing images of 'rarely seen' scenarios [8,9,10,2].…”

Section: % 50%mentioning

confidence: 99%

“…Following [8], most of the current state-of-the-art CS networks [8,2,10,9] use a unified training dataset, i.e., Train91 (see the first row in Fig. 5), for the convenience of comparison.…”

Section: Analysis On Robustness For Various Scenesmentioning

confidence: 99%

See 3 more Smart Citations

ISTA-NET⁺⁺: Flexible Deep Unfolding Network for Compressive Sensing

You

Xie

Zhang

2021

2021 IEEE International Conference on Multimedia and Expo (ICME)

View full text Add to dashboard Cite

While deep neural networks have achieved impressive success in image compressive sensing (CS), most of them lack flexibility when dealing with multi-ratio tasks and multiscene images in practical applications. To tackle these challenges, we propose a novel end-to-end flexible ISTAunfolding deep network, dubbed ISTA-Net ++ , with superior performance and strong flexibility. Specifically, by developing a dynamic unfolding strategy, our model enjoys the adaptability of handling CS problems with different ratios, i.e., multi-ratio tasks, through a single model. A cross-block strategy is further utilized to reduce blocking artifacts and enhance the CS recovery quality. Furthermore, we adopt a balanced dataset for training, which brings more robustness when reconstructing images of multiple scenes. Extensive experiments on four datasets show that ISTA-Net ++ achieves state-of-the-art results in terms of both quantitative metrics and visual quality. Considering its flexibility, effectiveness and practicability, our model is expected to serve as a suitable baseline in future CS research.

show abstract

Section: Comparison With State-of-the-art Methodsmentioning

confidence: 99%

Section: % 50%mentioning

confidence: 99%

Section: % 50%mentioning

confidence: 99%

“…Following [8], most of the current state-of-the-art CS networks [8,2,10,9] use a unified training dataset, i.e., Train91 (see the first row in Fig. 5), for the convenience of comparison.…”

Section: Analysis On Robustness For Various Scenesmentioning

confidence: 99%

See 2 more Smart Citations

ISTA-NET⁺⁺: Flexible Deep Unfolding Network for Compressive Sensing

You

Xie

Zhang

2021

2021 IEEE International Conference on Multimedia and Expo (ICME)

View full text Add to dashboard Cite

show abstract

“…Structure components indicates dominant structures of images, while texture components contains the information of the morphological details (see Figure 1). The texture component can well reveal the essential characteristic of an image and its information can help to reconstruct the image with compressive sensing methods [16]. The image texture component indeed is the high-frequency information, and many methods can be applied to achieve it.…”

Section: High-frequency Texture Componentmentioning

confidence: 99%

Spectral clustering based on high‐frequency texture components for face datasets

Liang

Guo

Liu

et al. 2021

IET Image Processing

View full text Add to dashboard Cite

Spectral clustering is one of the most widely used technologies for clustering tasks, which represents data as a weighted graph, and aims to find an appropriate way to cut the graph apart in order to categorize the raw data. The pivotal step of spectral clustering is to find out the accurate information to estimate the relationship of pairwise data, based on which a graph can be constructed. According to the cognition that different faces are distinguished by the edge contour which can be represented by high-frequency texture components, a novel spectral clustering algorithm via high-frequency signal of human face, named highfrequency spectral clustering (HFSC) is proposed. In HFSC, first the local high-frequency texture components are extracted from samples. Then the relationship of pairwise samples can be estimated with the degree of correlation, which is produced from the image high-frequency information. The graph will be set up with the correlation information. Subsequently the graph cut will be implemented to achieve the final clustering results. Experimental results show that this algorithm outperforms the state-of-the-art clustering methods on several datasets.This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

show abstract