PSD-ELGAN:基于伪自蒸的CycleGAN,具有增强的局部对抗性交互,用于单个图像的消毒
概括
这项研究引入了一种新的图像消毒方法,PSD-ELGAN,使用伪自蒸和增强的局部对抗学习. 它改善了无雾的特征表示,并有效地去除剩余的雾,以获得更好的图像质量.
科学领域:
- 计算机视觉 计算机视觉
- 人工智能的人工智能
- 图像处理 图像处理
背景情况:
- 现有的基于CycleGAN的除雾方法由于软约束而难以准确地建模无雾特征.
- 全球区分器可能错误地分类深度变化的模糊图像,导致剩余的模糊.
研究的目的:
- 提出一种新的图像除尘算法,PSD-ELGAN,可以克服当前方法的局限性.
- 为了增强没有雾的特征表示,并改善剩余雾抑制.
主要方法:
- 使用CycleGAN在无监督框架中生成伪图像对.
- 采用伪自蒸用于从清洁到模糊的图像生成器转移知识.
- 实施增强的局部对抗性相互作用,通过适应性补丁选择来消除残留的雾.
主要成果:
- 在没有增加网络参数的情况下,PSD-ELGAN展示了改进的无雾特征表示.
- 该方法有效地抑制了挑战性地区的残留雾.
- 实验结果显示,在各种数据集中,性能和通用性都很有希望.
结论:
- 通过结合伪自蒸和局部对抗性学习,PSD-ELGAN提供了一种强大的图像消毒解决方案.
- 拟议的方法提高了脱图像的质量和清晰度.
更多相关视频
09:00Indoor Experimental Assessment of the Efficiency and Irradiance Spot of the Achromatic Doublet on Glass ADG Fresnel Lens for Concentrating Photovoltaics
Published on: October 27, 2017
8.8K
11:24Evaporation-reducing Culture Condition Increases the Reproducibility of Multicellular Spheroid Formation in Microtiter Plates
Published on: March 7, 2017
7.0K
相关概念视频
![[4+2] Cycloaddition of Conjugated Dienes: Diels–Alder Reaction](/_next/image?url=https%3A%2F%2Fcloudfront.jove.com%2Ffiles%2Fmedia%2Fscience-education%2Fscience-education-thumbs%2F12417.jpg&w=3840&q=75)
01:16[4+2] Cycloaddition of Conjugated Dienes: Diels–Alder Reaction
10.1K
The Diels–Alder reaction is an example of a thermal pericyclic reaction between a conjugated diene and an alkene or alkyne, commonly referred to as a dienophile. The reaction involves a concerted movement of six π electrons, four from the diene and two from the dienophile, forming an unsaturated six-membered ring. As a result, these reactions are classified as [4+2] cycloadditions.
10.1K
01:12Cycloaddition Reactions: MO Requirements for Photochemical Activation
2.0K
Some cycloaddition reactions are activated by heat, while others are initiated by light. For example, a [2 + 2] cycloaddition between two ethylene molecules occurs only in the presence of light. It is photochemically allowed but thermally forbidden.
2.0K