Joint operation and attention block search for lightweight image restoration

Article Computer Science, Artificial Intelligence

Semi-supervised student-teacher learning for single image super-resolution

Lin Wang et al.

Summary: In this paper, a novel Semi-supervised Student-Teacher Super-Resolution approach is proposed to address the challenge of limited labeled LR images and degradation kernels in single image super-resolution (SISR). The approach combines knowledge transfer, partial weight sharing, and a pair matching network to effectively utilize both labeled and unlabeled LR images for super-resolution. Experimental results demonstrate that the proposed method substantially outperforms unsupervised methods and performs favorably over fully supervised methods.

PATTERN RECOGNITION (2022)

Add to Collection

Article Computer Science, Artificial Intelligence

You Only Look Yourself: Unsupervised and Untrained Single Image Dehazing Neural Network

Boyun Li et al.

Summary: This paper proposes an unsupervised and untrained neural network model YOLY for single image dehazing, which separates and composes the observed hazy image through layer disentanglement, bypassing the traditional training paradigm to achieve promising performance.

INTERNATIONAL JOURNAL OF COMPUTER VISION (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

Image denoising using complex-valued deep CNN

Yuhui Quan et al.

Summary: This study investigates the potentials of complex-valued convolutional neural networks for image denoising. Experimental results show competitive performance of the proposed complex-valued denoising CNN against existing state-of-the-art real-valued denoising CNNs, with better robustness to possible inconsistencies of noise models. Complex-valued CNNs provide another promising deep-learning-based approach to image denoising and other image recovery tasks.

PATTERN RECOGNITION (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

Image super -resolution via channel attention and spatial graph convolutional network

Yue Yang et al.

Summary: In this work, a channel attention and spatial graph convolutional network (CASGCN) is proposed for more powerful feature obtaining and feature correlations modeling by incorporating global spatial and channel inter-dependencies. CASGCN utilizes channel and spatial branches to learn the representation of each image pixel, using attention mechanism and spatial-aware graph to extract informative features and model global self-similar information. The adjacency matrix in spatial-aware graph is dynamically generated via the Gram matrix to model global correlations between pixels and is shared across the whole network without auxiliary parameters. Extensive experiments demonstrate the effectiveness of CASGCN in single image superresolution tasks.

PATTERN RECOGNITION (2021)

Add to Collection

Article Computer Science, Artificial Intelligence

Residual Dense Network for Image Restoration

Yulun Zhang et al.

Summary: This work proposes a novel and efficient residual dense network (RDN) to address the problem of relatively-low performance in image restoration by better balancing efficiency and effectiveness in utilizing hierarchical features from all convolutional layers. The RDN utilizes residual dense blocks (RDB) and local feature fusion to extract abundant local features and adaptively learn more effective features, while also employing global feature fusion for learning global hierarchical features in a holistic way. Experimental results demonstrate that the RDN outperforms state-of-the-art methods in various image restoration tasks both quantitatively and visually.

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE (2021)

Add to Collection

Article Computer Science, Artificial Intelligence