Filters








295 Hits in 6.3 sec

Robust Super-Resolution GAN, with Manifold-based and Perception Loss [article]

Uddeshya Upadhyay, Suyash P. Awate
2019 arXiv   pre-print
We propose a loss based on an autoencoder-based manifold-distance between the super-resolved and high-resolution images, to reproduce realistic textural content in super-resolved images.  ...  We propose to learn to super-resolve images to match human perceptions of structure, luminance, and contrast.  ...  We propose a novel GAN-based learning framework for super-resolution that is robust to errors in training-set curation, using quasi-norm based loss functions.  ... 
arXiv:1903.06920v1 fatcat:q53u5tgy5jbhtftadtsdianwdi

Natural and Realistic Single Image Super-Resolution with Explicit Natural Manifold Discrimination [article]

Jae Woong Soh, Gu Yong Park, Junho Jo, Nam Ik Cho
2019 arXiv   pre-print
Our results show better naturalness compared to the recent super-resolution algorithms including perception-oriented ones.  ...  However, the networks trained with objective loss functions generally fail to reconstruct the realistic fine textures and details that are essential for better perceptual quality.  ...  Natural and Realistic Super-Resolution In this section, we explain the proposed natural and realistic super-resolution (NatSR) generator model and the training loss function.  ... 
arXiv:1911.03624v1 fatcat:4axaj7zrhrdxvmikvgdpobdyam

Natural and Realistic Single Image Super-Resolution With Explicit Natural Manifold Discrimination

Jae Woong Soh, Gu Yong Park, Junho Jo, Nam Ik Cho
2019 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
Our results show better naturalness compared to the recent super-resolution algorithms including perception-oriented ones.  ...  However, the networks trained with objective loss functions generally fail to reconstruct the realistic fine textures and details that are essential for better perceptual quality.  ...  Natural and Realistic Super-Resolution In this section, we explain the proposed natural and realistic super-resolution (NatSR) generator model and the training loss function.  ... 
doi:10.1109/cvpr.2019.00831 dblp:conf/cvpr/SohPJC19 fatcat:nnh34jnkmfdjhb7tkmgoahauji

A Latent Encoder Coupled Generative Adversarial Network (LE-GAN) for Efficient Hyperspectral Image Super-resolution [article]

Yue Shi, Liangxiu Han, Lianghao Han, Sheng Chang, Tongle Hu, Darren Dancey
2021 arXiv   pre-print
AVIRIS and UHD-185) for various upscaling factors and added noise levels, and compared with the state-of-the-art super-resolution models (i.e. HyCoNet, LTTR, BAGAN, SR- GAN, WGAN).  ...  Realistic hyperspectral image (HSI) super-resolution (SR) techniques aim to generate a high-resolution (HR) HSI with higher spectral and spatial fidelity from its low-resolution (LR) counterpart.  ...  Strategy (BEIS)) and the Open Research Fund of Key Laboratory of Digital Earth Science, Chinese Academy of Sciences(No.2019LDE003).  ... 
arXiv:2111.08685v1 fatcat:xfabty4425f2ljiwdjpwyonxd4

Perception-Enhanced Image Super-Resolution via Relativistic Generative Adversarial Networks [chapter]

Thang Vu, Tung M. Luu, Chang D. Yoo
2019 Lecture Notes in Computer Science  
This paper considers a deep Generative Adversarial Networks (GAN) based method referred to as the Perception-Enhanced Super-Resolution (PESR) for Single Image Super Resolution (SISR) that enhances the  ...  perceptual quality of the reconstructed images by considering the following three issues: (1) ease GAN training by replacing an absolute with a relativistic discriminator, (2) include in the loss function  ...  Conclusion We have presented a deep Generative Adversarial Network (GAN) based method referred to as the Perception-Enhanced Super-Resolution (PESR) for Single Image Super Resolution (SISR) that enhances  ... 
doi:10.1007/978-3-030-11021-5_7 fatcat:v6mizuzxzrctnoize3xcpf7d4i

Manifold Matching via Deep Metric Learning for Generative Modeling [article]

Mengyu Dai, Haibin Hang
2021 arXiv   pre-print
; in super-resolution task, we incorporate the framework in perception-based models and improve visual qualities by producing samples with more natural textures.  ...  It is achieved by matching two sets of points using their geometric shape descriptors, such as centroid and p-diameter, with learned distance metric; the metric generator utilizes both real data and generated  ...  Final results with both distortion-based and perception-based evaluation scores on benchmark datasets are presented in Table 5 , where GAN-ResNet represents SRGAN without VGG component in loss function  ... 
arXiv:2106.10777v3 fatcat:eabulpgy6bcexjfmcivhin2pia

Best-Buddy GANs for Highly Detailed Image Super-Resolution [article]

Wenbo Li, Kun Zhou, Lu Qi, Liying Lu, Nianjuan Jiang, Jiangbo Lu, Jiaya Jia
2021 arXiv   pre-print
We consider the single image super-resolution (SISR) problem, where a high-resolution (HR) image is generated based on a low-resolution (LR) input.  ...  An ultra-high-resolution 4K dataset is also constructed to facilitate future super-resolution research.  ...  Ground-truth-based LPIPS is more representative and robust than no-reference-based PI. Zoom in for better visual comparison.  ... 
arXiv:2103.15295v3 fatcat:xljujqfacjdevnmgn5vdrdq764

Creating High Resolution Images with a Latent Adversarial Generator [article]

David Berthelot, Peyman Milanfar, Ian Goodfellow
2020 arXiv   pre-print
Single image super-resolution is the task of predicting the image closest to the ground truth from a relatively low resolution image.  ...  We propose to produce samples of high resolution images given extremely small inputs with a new method called Latent Adversarial Generator (LAG).  ...  In the context of super-resolution from a low resolution (LR) input, this task has been researched extensively and the current best methods are based on deep learning (DL).  ... 
arXiv:2003.02365v1 fatcat:w33o32dwobbzlo3jltvqrok25u

Perceptual Extreme Super Resolution Network with Receptive Field Block [article]

Taizhang Shang, Qiuju Dai, Shengchen Zhu, Tong Yang, Yandong Guo
2020 arXiv   pre-print
To tackle this difficulty, we develop a super resolution network with receptive field block based on Enhanced SRGAN. We call our network RFB-ESRGAN. The key contributions are listed as follows.  ...  First, for the purpose of extracting multi-scale information and enhance the feature discriminability, we applied receptive field block (RFB) to super resolution.  ...  We also attempt to fuse the models with more GAN-based models. For instance, use 20 or 40 best GAN-based models for ensemble.  ... 
arXiv:2005.12597v1 fatcat:4tjretl65nc2hm5uzx5tua652u

One-to-many Approach for Improving Super-Resolution [article]

Sieun Park, Eunho Lee
2021 arXiv   pre-print
In this paper, we adapt the concepts of SRFlow to improve GAN-based super-resolution by properly implementing the one-to-many property.  ...  Recently, there has been discussions on the ill-posed nature of super-resolution that multiple possible reconstructions exist for a given low-resolution image.  ...  We believe the current GAN-based solutions for super-resolution is missing key components of a one-to-many pipeline.  ... 
arXiv:2106.10437v4 fatcat:ezbeaptaqjbtjh32wdrkqlxgze

Advanced Single Image Resolution Upsurging Using A Generative Adversarial Network

Md. Moshiur Rahman, Samrat Kumar Dey, Kabid Hassan Shibly
2020 Signal & Image Processing An International Journal  
A higher resolution of an image is important for various fields such as medical imaging; astronomy works and so on as images of lower resolution becomes unclear and indistinct when their sizes are enlarged  ...  In this paper, we have proposed a technique of generating higher resolution images form lower resolution using Residual in Residual Dense Block network architecture with a deep network.  ...  Here we use loss function in (3) . Pre-training with loss based on pixel helps GAN-based methods to gain visually better results.  ... 
doi:10.5121/sipij.2020.11105 fatcat:cn2cifwd5bcsfezjeopmet47ru

Reference based Face Super-resolution

Zhi-Song Liu, Wan-Chi Siu, Yui-Lam Chan
2019 IEEE Access  
We create a benchmark dataset on reference based face super-resolution (RefSR-Face) for general research use, which contains reference images paired with low-resolution images of various pose, emotions  ...  We propose a novel Conditional Variational AutoEncoder model for this Reference based Face Super-Resolution (RefSR-VAE).  ...  Note that FSRGAN and FSRNet are the results from [35] . FSRGAN is the GAN version of FSRNet that was trained based on perception loss rather than pixel based MSE loss.  ... 
doi:10.1109/access.2019.2934078 fatcat:qw6mh56ysvfpjbvno4hxn7sgwq

SRPGAN: Perceptual Generative Adversarial Network for Single Image Super Resolution [article]

Bingzhe Wu, Haodong Duan, Zhichao Liu, Guangyu Sun
2017 arXiv   pre-print
In the framework, we propose a robust perceptual loss based on the discriminator of the built SRPGAN model.  ...  We use the Charbonnier loss function to build the content loss and combine it with the proposed perceptual loss and the adversarial loss.  ...  We can get a more robust perceptual loss for image super resolution by that.  ... 
arXiv:1712.05927v2 fatcat:b6bfrhmp3zflfkjprv75xuyrme

HRINet: Alternative Supervision Network for High-resolution CT image Interpolation [article]

Jiawei Li, Jae Chul Koh, Won-Sook Lee
2020 arXiv   pre-print
We compare an MSE based and a perceptual based loss optimizing methods for high quality interpolation, and show the tradeoff between the structural correctness and sharpness.  ...  We combine the idea of ACAI and GANs, and propose a novel idea of alternative supervision method by applying supervised and unsupervised training alternatively to raise the accuracy of human organ structures  ...  [2] , image super-resolution [3] [4] and neural style transfer [5] [6], etc.  ... 
arXiv:2002.04455v2 fatcat:6hgt7uab7vhcrkglzrxaevjwdy

2021 Index IEEE Transactions on Image Processing Vol. 30

2021 IEEE Transactions on Image Processing  
The primary entry includes the coauthors' names, the title of the paper or other item, and its location, specified by the publication abbreviation, year, month, and inclusive pagination.  ...  The Subject Index contains entries describing the item under all appropriate subject headings, plus the first author's name, the publication abbreviation, month, and year, and inclusive pages.  ...  ., +, TIP 2021 9270-9279 VolumeNet: A Lightweight Parallel Network for Super-Resolution of MR Multi-Stream Fusion Network With Generalized Smooth L 1 Loss for Single and CT Volumetric Data.  ... 
doi:10.1109/tip.2022.3142569 fatcat:z26yhwuecbgrnb2czhwjlf73qu
« Previous Showing results 1 — 15 out of 295 results