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

Chen, Yunchun; Gao, Chen; Robb, Esther; Huang, Jia-Bin

doi:10.48550/arxiv.2008.11713

Cited by 3 publications

(4 citation statements)

References 59 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, we would like to emphasize that the default architecture of DIP is employed just for justifying the function of robust statistics. Surely, it is possible to further improve the proposed algorithm by using the neural architecture search reported in [50].…”

Section: Discussionmentioning

confidence: 99%

Unsupervised Hyperspectral Denoising Based on Deep Image Prior and Least Favorable Distribution

Niresi

Chi

2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

This paper considers the inverse problem under hyperspectral images (HSIs) denoising framework. Recently, it has been shown that deep learning is a promising approach to image denoising. However, deep learning to be effective usually needs a massive amount of training data. Moreover, in a practical scenario, HSIs may get contaminated by different kinds of noises such as Gaussian and/or sparse noise. Lately, it has been reported that the convolutional neural network (CNN), the core element used by deep image prior (DIP), is able to capture image statistical characteristics without the need of training, i.e., restore the clean image blindly. Nonetheless, there exists some performance gap between DIP and state-of-the-art methods in HSIs (e.g., low-rank models). By applying the Huber loss function (HLF), which is derived through a least favorable distribution in robust statistics, to DIP, we propose a novel unsupervised denoising algorithm, referred as to the HLF-DIP, free from pre-training and without involving any regularizer. Extensive experimental results are provided to demonstrate that the proposed HLF-DIP algorithm significantly outperforms seven state-of-the-art algorithms in both complexity (thanks to no regularization) and robustness against complex noise (e.g., mixed types of noises).

show abstract

Section: Discussionmentioning

confidence: 99%

Unsupervised Hyperspectral Denoising Based on Deep Image Prior and Least Favorable Distribution

Niresi

Chi

2022

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

show abstract

“…These optimized network architectures have been shown to enhance the performance of classic UNNP methods. Instead of using hand-crafted neural networks for UNNP, Chen et al [69] proposed the use of deep reinforcement learning (DRL) to search for the best possible neural network architecture for a specific problem. Their work is inspired by the NAS algorithms [77], [78], [79], which involve the search of optimal neural networks that give the top performance on large datasets.…”

Section: Neural Architecture Search For Unnpmentioning

confidence: 99%

“…4: Different UNNP architectures proposed in the literature. Relevant papers: (a) [6] ; (b) [12] ; (c) [7], [25], [49] ; (d) [37], [68], [69] ; (e) [28], [70] ; (f) [67] ; (g) [39]. Fig.…”

Section: Untrained Neural Network Priors: An Introductionmentioning

confidence: 99%

Untrained Neural Network Priors for Inverse Imaging Problems: A Survey

Qayyum

Ilahi

Shamshad

et al. 2022

IEEE Trans. Pattern Anal. Mach. Intell.

View full text Add to dashboard Cite

In recent years, advancements in machine learning (ML) techniques, in particular, deep learning (DL) methods have gained a lot of momentum in solving inverse imaging problems, often surpassing the performance provided by hand-crafted approaches. Traditionally, analytical methods have been used to solve inverse imaging problems such as image restoration, inpainting, and superresolution. Unlike analytical methods for which the problem is explicitly defined and the domain knowledge is carefully engineered into the solution, DL models do not benefit from such prior knowledge and instead make use of large datasets to predict an unknown solution to the inverse problem. Recently, a new paradigm of training deep models using a single image, named untrained neural network prior (UNNP) has been proposed to solve a variety of inverse tasks, e.g., restoration and inpainting. Since then, many researchers have proposed various applications and variants of UNNP. In this paper, we present a comprehensive review of such studies and various UNNP applications for different tasks and highlight various open research problems which require further research.

show abstract

“…The output at the first iteration contains randomization due to random code vector z and the optimizer iteratively optimizes neural network's parameters θ for the given input image using backpropagation. It has been shown in the literature [19], [36] that the choice of neural network architectures has a direct impact on the performance, e.g., we can design/handcraft a particular neural network architecture for modeling a specific image [21]. This serves as a solution space when modeling images using untrained neural networks priors.…”

Section: Untrained and Pretrained Cdipsmentioning

confidence: 99%