Deep Residual Network for Image Super-Resolution Reconstruction

Zhang, Bozhen; Gao, Cheng

doi:10.1109/cyber55403.2022.9907617

Cited by 3 publications

(2 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We conclude some potential challenges and limitations in OW-CZSL, e.g., the limited semantic relations between shoes and materials on UT-Zappos, the difficult state recognition on MIT-States and C-GQA, and the difficulty in learning the annotation priority of humans. In the future, we plan to implement dynamic γ selections and enhance the network structure of the backbone [66], [67], which may fix the learning deficiency and balance the learning tendencies dynamically. In this way, our model, SAD-SP, may be capable of addressing the challenges found in our experiments, especially when tackling datasets with large numbers of simple primitives.…”

Section: Discussionmentioning

confidence: 99%

Simple Primitives with Feasibility- and Contextuality-Dependence for Open-World Compositional Zero-shot Learning

Liu¹,

Li²,

Yao³

et al. 2022

Preprint

View full text Add to dashboard Cite

The task of Compositional Zero-Shot Learning (CZSL) is to recognize images of novel state-object compositions that are absent during the training stage. Previous methods of learning compositional embedding have shown effectiveness in closed-world CZSL. However, in Open-World CZSL (OW-CZSL), their performance tends to degrade significantly due to the large cardinality of possible compositions. Some recent works separately predict simple primitives (i.e., states and objects) to reduce cardinality. However, they consider simple primitives as independent probability distributions, ignoring the heavy dependence between states, objects, and compositions. In this paper, we model the dependence of compositions via feasibility and contextuality. Feasibility-dependence refers to the unequal feasibility relations between simple primitives, e.g., hairy is more feasible with cat than with building in the real world. Contextuality-dependence represents the contextual variance in images, e.g., cat shows diverse appearances under the state of dry and wet. We design Semantic Attention (SA) and generative Knowledge Disentanglement (KD) to learn the dependence of feasibility and contextuality, respectively. SA captures semantics in compositions to alleviate impossible predictions, driven by the visual similarity between simple primitives. KD disentangles images into unbiased feature representations, easing contextual bias in predictions. Moreover, we complement the current compositional probability model with feasibility and contextuality in a compatible format. Finally, we conduct comprehensive experiments to analyze and validate the superior or competitive performance of our model, Semantic Attention and knowledge Disentanglement guided Simple Primitives (SAD-SP), on three widely-used benchmark OW-CZSL datasets.

show abstract

Section: Discussionmentioning

confidence: 99%

Simple Primitives with Feasibility- and Contextuality-Dependence for Open-World Compositional Zero-shot Learning

Liu¹,

Li²,

Yao³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…As the existing SR models based on convolutional neural networks mainly focus on designing deepened or widened networks [9], [10] while ignoring the loss of high-frequency feature information of images, scholars have proposed a series of solutions to this problem. For example, by using convolution kernels with different scales of 1 × 1, 3 × 3 and 5 × 5 to extract rich feature information, Zhang and Guo [11] obtained reconstructed images with better subjective visual effects and objective evaluation indexes. By combining multi-scale convolution kernel with residual structure, Lu et al [12] proposed a multi-scale information aggregation block to extract image features efficiently without increasing the number of parameters.…”

Section: Introductionmentioning

confidence: 99%

Image Super-Resolution Based on Residual Attention and Multi-Scale Feature Fusion

et al. 2023

View full text Add to dashboard Cite

At present, deep residual network has been widely used in image super-resolution and proved to be able to achieve good reconstruction results. However, the existing super-resolution algorithms based on deep residual network have the problems of indiscriminately learning feature information of different regions and low utilization rate of feature information, which make them difficult to further improve the reconstruction effect. In view of the above problems, a novel super-resolution reconstruction network based on residual attention and multi-scale feature fusion (RAMF) is proposed in this paper. Firstly, a lightweight multi-scale residual module (LMRM) is proposed in the deep feature extraction stage, by which the multi-scale features are extracted and further cross-connected to enrich the information of different receptive fields. Then, to fully improve the utilization rate of feature information, a dense feature fusion structure is designed to fuse the output feature of each LMRM. Finally, a residual spatial attention module (RSAM) is proposed to specifically learn and better retain high-frequency feature information, so as to improve the reconstruction effect. Experimental tests and comparisons are conducted with the current advanced methods on four baseline databases, and the results demonstrate that the proposed RAMF can achieve better reconstruction effect with fewer parameters, low computational complexity, fast processing speed and high objective evaluation index. Especially, the peak signal-to-noise ratio measured on Urban100 data set increases by 0.13dB on average, and the reconstructed image has better visual effect and richer texture detail features.

show abstract

Simple Primitives With Feasibility- and Contextuality-Dependence for Open-World Compositional Zero-Shot Learning

Liu,

Li,

Yao

et al. 2024

IEEE Trans. Pattern Anal. Mach. Intell.

View full text Add to dashboard Cite

Deep Residual Network for Image Super-Resolution Reconstruction

Cited by 3 publications

References 13 publications

Simple Primitives with Feasibility- and Contextuality-Dependence for Open-World Compositional Zero-shot Learning

Simple Primitives with Feasibility- and Contextuality-Dependence for Open-World Compositional Zero-shot Learning

Image Super-Resolution Based on Residual Attention and Multi-Scale Feature Fusion

Simple Primitives With Feasibility- and Contextuality-Dependence for Open-World Compositional Zero-Shot Learning

Contact Info

Product

Resources

About