Light Weight IBP Deep Residual Network for Image Super Resolution

Lin, Hai; Yang, Junjie

doi:10.1109/access.2021.3091899

Cited by 6 publications

(5 citation statements)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…During training, we set the batchsize to B, and first sample B random scales r 1∼B in uniform distribution U (1,4). Then we crop B pathches with size {48r i × 48r i } B i=1 from training images.…”

Section: ) Implementation Detailsmentioning

confidence: 99%

See 1 more Smart Citation

Feature Preserving and Enhancing Network for Image Super-Resolution

Su,

Li,

et al. 2023

IEEE Access

View full text Add to dashboard Cite

Single image super-resolution (SISR) with deep convolutional neural networks has recently attracted increasing attention due to its potentials to generate rich details. To obtain better fidelity and visual quality, most of existing methods are of heavy design with the depth of network. However, the lack of highfrequency information in the deep network and the limit of non-local context relations of pixels remain the core issue. In addition, there are many arbitrary-scale image super-resolution tasks, so it is a focus that provides an effective and efficient model for super resolution with arbitrary-scale. To this end, we propose a Feature Preserving and Enhancing Network (FPEN) based on implicit representation, which is aims to preserving high-frequency information in deep network and enhancing non-local contextual features of pixels to output more realistic images at arbitrary scales. In particular, our proposed High Frequency Preserving Block (HFPB) can divide the features into high-frequency components and low-frequency components, and allocate more operations to high-frequency components to ensure that high-frequency components can be preserved in the network. Since high-frequency information contains the details and texture of the image, it can restore the finer details of the image. Moreover, Pixel Continuity Attention (PiCA) module our proposed utilizes visual cues observed from pixels to adaptively recalibrate the pixel range that needs attention in the image to better achieve feature enhancement and generate smoother images. Extensive experiments conducted on benchmark SISR models and datasets show that Feature Preserving and Enhancing Network can be employed for various SISR tasks with arbitrary-scale to obtain the better visual quality than other state-of-the-art SR algorithms.INDEX TERMS Image super resolution with arbitrary-scale, high frequency preservingsingle methods, non-local attention methods, implicit neural representation.

show abstract

Section: ) Implementation Detailsmentioning

confidence: 99%

“…S INGLE image super resolution reconstruction (SISR) is an important research topic [1], whose goal is to reconstruct high-resolution images from low-resolution images as shown in Figure 1. SISR has been widely used in many fields, including medical imaging [2], image compression [3], small object detection [4], etc.…”

Section: Introductionmentioning

confidence: 99%

Feature Preserving and Enhancing Network for Image Super-Resolution

Su,

Li,

et al. 2023

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Our work is related to the recent design of tiny and low-cost neural networks [25]- [28]. These studies propose methods that have the potential to enhance the application of lightweight neural networks.…”

Section: Related Workmentioning

confidence: 99%

Accelerating Super-Resolution Network Inference via Sensitivity-Based Weight Sparsity Allocation

Nguyen,

Lee

et al. 2023

IEEE Access

View full text Add to dashboard Cite

Weight sparsification has been extensively studied in image classification and object detection to accelerate network inference. However, for image generation tasks, such as image super-resolution, forcing some weights to zeros is a non-trivial task that typically causes significant degradation in restoration quality, that is, peak signal-to-noise (PSNR). In this study, we first introduce a sensitivity metric to measure PSNR degradation for layer-wise sparsity changes and observe that the sensitivities vary significantly across network layers. We demonstrate that a uniform sparsity allocation method generally causes a nonnegligible decrease in accuracy, that is, approximately 0.17 dB, for 65% of the zero weights. In addition, finding an optimal solution to the sparsity allocation problem is not feasible because the design space is exponential with respect to the number of weights and layers. To address this problem, we proposed a simple yet effective sparsity allocation method based on layer-wise sensitivity. The experimental results demonstrate that the proposed method achieves up to 35% computation reduction with an average accuracy drop of 0.02 dB varying between 0.01 to 0.04 dB across the well-known datasets Set5, Set14, B100, and Urban100. Moreover, when integrated with the activation sparsity SMSR, the proposed method reduced the computation by 46% on average.

show abstract

“…In order to better analyze the calculation results of the residual neural network, different analysis methods are adopted to analyze the deep residual network [19,20].…”

Section: Evaluation Of Residual Networkmentioning

confidence: 99%

Leisure Sports Behavior Recognition Algorithm Based on Deep Residual Network

Chou

2022

Scientific Programming

View full text Add to dashboard Cite

Aiming at the related problems existing in the field of leisure sports computing, in order to study the behavior recognition of leisure sports by deep residual network, based on the deep residual neural network theory, the behavior recognition algorithm and the corresponding robust model are used to analyze the leisure sports related samples, and the correlation model is used to predict and analyze the leisure sports related content. The results show that the change curves of Sig and Tanh functions can be divided into slow increasing stage, linear increasing stage, and stable stage. The y value corresponding to ReLU curve shows a linear change trend with the increase of x value. The Leaky function’s corresponding curve can be divided into two stages. The function coincides with the ReLU function in the first quadrant and remains linear in the third quadrant. The activation function curves corresponding to layers 56 and 20 have a relatively large variation range, and both of them show an overall trend of gradual decline. On the whole, the curve value corresponding to layer 56 is higher than that corresponding to layer 20, indicating that the method of layer 20 is relatively good and the corresponding training error is relatively low. It can be seen from the robustness recognition rate of various methods under different training samples that Fl has the highest overall data recognition rate while Sc has relatively poor stability. However, the recognition rate of IDCC and DCC shows a relatively flat trend, indicating that these two methods have certain advantages in describing the robust recognition rate. The research results can provide theoretical support for the application of deep residual neural networks in other fields.

show abstract

Light Weight IBP Deep Residual Network for Image Super Resolution

Cited by 6 publications

References 40 publications

Feature Preserving and Enhancing Network for Image Super-Resolution

Feature Preserving and Enhancing Network for Image Super-Resolution

Accelerating Super-Resolution Network Inference via Sensitivity-Based Weight Sparsity Allocation

Leisure Sports Behavior Recognition Algorithm Based on Deep Residual Network

Contact Info

Product

Resources

About