Designing a Practical Degradation Model for Deep Blind Image Super-Resolution

Zhang, Kai; Liang, Jingyun; Gool, Luc Van; Timofte, Radu

doi:10.1109/iccv48922.2021.00475

Cited by 259 publications

(163 citation statements)

References 50 publications

Supporting

Mentioning

163

Contrasting

Order By: Relevance

“…We compare the proposed QuanTexSR with several state-of-the-art methods for blind SR, including CDC [45], DAN [29], DASR(W) [40], RealSR [17], BSRGAN [51], Real-ESRGAN+ [42] and SwinIR-GAN [26]. Specifically, CDC proposed a divide-and-conquer architecture; DAN, DASR(W) and RealSR learned degradation models from LR inputs; BSRGAN, Real-ESRGAN+ and SwinIR-GAN used synthetic training data generated by handcrafted degradation models.…”

Section: Comparison With Existing Methodsmentioning

confidence: 99%

“…Note that we focus on the LPIPS scores as it better captures the perceptual quality than other metrics (e.g. PSNR/SSIM) [41,42,48,51,54]. In addition, it can be observed that: in general, methods that learn the degradations, such as DAN and DASR(W), perform much worse than those using manually designed degradation models, which indicates the difficulties in learning complex real-world degradations.…”

Section: Comparison With Existing Methodsmentioning

confidence: 99%

“…On the contrary, explicit methods aim to synthesize "real" LR images by a manually designed degradation process. Specifically, BSRGAN [51] and Real-ESRGAN [42] describe different ways to improve the common image degradation pipeline. Both of them demonstrate much better visual quality than implicit methods in blind SISR.…”

Section: Related Workmentioning

confidence: 99%

“…Training Dataset We follow BSRGAN [51] and build a training set that includes DIV2K [1], Flickr2K [27], DIV8K [13] and 10,000 face images from FFHQ [19]. Evaluation Metrics For synthetic test datasets with ground truth images, we employ the well-known perceptual metric, LPIPS [54] score, to evaluate the perceptual quality of generated images.…”

Section: Datasets and Evaluation Metricsmentioning

confidence: 99%

“…In addition, it can be observed that different combinations of code samples correspond to different semantics, such as, 1 Table 1. Quantitative comparison with state-of-the-art methods on synthetic benchmarks for blind SR. LR images are generated with a mixed degradation model of BSRGAN [51] and Real-ESRGAN [42]. PSNR/SSIM ↑: the higher, the better; LPIPS ↓: the lower, the better.…”

Section: Analysis Of Texture Priorsmentioning

confidence: 99%

See 4 more Smart Citations

Real-World Blind Super-Resolution via Feature Matching with Implicit High-Resolution Priors

Chen¹,

Shi²,

Qin³

et al. 2022

Preprint

View full text Add to dashboard Cite

A key challenge of blind image super resolution is to recover realistic textures for low-resolution images with unknown degradations. Most recent works completely rely on the generative ability of GANs, which are difficult to train. Other methods resort to high-resolution image references that are usually not available. In this work, we propose a novel framework, denoted as QuanTexSR, to restore realistic textures with the Quantized Texture Priors encoded in Vector Quantized GAN. The QuanTexSR generates textures by aligning the textureless content features to the quantized feature vectors, i.e., a pretrained feature codebook. Specifically, QuanTexSR formulates the texture generation as a feature matching problem between textureless features and a pretrained feature codebook. The final textures are then generated by the quantized features from the codebook. Since features in the codebook have shown the ability to generate natural textures in the pretrain stage, QuanTexSR can generate rich and realistic textures with the pretrained codebook as texture priors. Moreover, we propose a semantic regularization technique that regularizes the pre-training of the codebook using clusters of features extracted from the pretrained VGG19 network. This further improves texture generation with semantic context. Experiments demonstrate that the proposed QuanTexSR can generate competitive or better textures than previous approaches.

show abstract

Section: Comparison With Existing Methodsmentioning

confidence: 99%

Section: Comparison With Existing Methodsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Datasets and Evaluation Metricsmentioning

confidence: 99%

Section: Analysis Of Texture Priorsmentioning

confidence: 99%

See 3 more Smart Citations

Real-World Blind Super-Resolution via Feature Matching with Implicit High-Resolution Priors

Chen¹,

Shi²,

Qin³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

P‐8.5: An Efficient Algorithm for Single Image Super‐Resolution

Liao,

Wu,

Lin

2023

Symp Digest of Tech Papers

View full text Add to dashboard Cite

In order to solve the problem of insufficient key information and serious damage of detail texture in the feature extraction process of single image super‐resolution reconstruction algorithm when processing low‐resolution image, we propose a high‐performance feature information dependence algorithm (HPDA) based on a Prior Preserving Block (PPB) and Residual Feature Holding Block (RFHB) to obtain feature information dependence between image blocks. To better focus key information in the original image, We introduced an improved Enhanced Spatial Attention module (ESA) to ensure the overall integrity and accuracy of the image reconstruction. The reconstruction quality detail and reconstruction speed of different scales have achieved a good balance.

show abstract

Metric Learning Based Interactive Modulation for Real-World Super-Resolution

Mou

Wang

et al. 2022

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Designing a Practical Degradation Model for Deep Blind Image Super-Resolution

Cited by 259 publications

References 50 publications

Real-World Blind Super-Resolution via Feature Matching with Implicit High-Resolution Priors

Real-World Blind Super-Resolution via Feature Matching with Implicit High-Resolution Priors

P‐8.5: An Efficient Algorithm for Single Image Super‐Resolution

Metric Learning Based Interactive Modulation for Real-World Super-Resolution

Contact Info

Product

Resources

About