A feature-supervised generative adversarial network for environmental monitoring during hazy days

Key Knowledgeable：

pair of hazy and clean images
使用一对晴雾天气的数据作为输入（用晴朗天气模拟雾天气or用雾天气模拟晴朗天气获得成对数据：本身二者就与真实的有差距，很大程度上取决于模拟的好坏），二者在generator的encoder部分各自编码获得特征后根据feature regularization loss进行监督约束.
假设任何一对模糊和干净的图像在共享的潜在空间中具有相似的信息，用于监督训练过程，以得到更好的结果。
Multi-scale discriminator
同时训练三个discriminator分别接受不同分辨率的输入,形成多尺度的识别.
Mutli-loss
Generator的loss由an adversarial loss, a style loss, a feature regularization loss and a perceptual loss四部分加权组成:
分别是:
1. Adversarial loss
  正常的generator loss:
2. Perceptual loss
  利用训练好的VGG网络对每层的特征进行监督:
3. Style loss
  解决了在风格上存在差异的问题:Gk为∅k矩阵变换得来
  2和3结合使用尽可能避免一些”人工产生”的模糊的区域和伪影.
4. Feature regularization loss
  约束pair input的特征关系:
- 正常的discriminator loss,只是将3个discriminator的loss进行相加做计算.

A feature-supervised generative adversarial network for environmental monitoring during hazy days相关推荐

Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network论文翻译——中英文对照
文章作者:Tyan 博客:noahsnail.com | CSDN | 简书声明:作者翻译论文仅为学习,如有侵权请联系作者删除博文,谢谢! 翻译论文汇总:https://github.com ...
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network Ledig C, Theis ...
【论文翻译】中英对照翻译--（Attentive Generative Adversarial Network for Raindrop Removal from A Single Image）
[开始时间]2018.10.08 [完成时间]2018.10.09 [论文翻译]Attentive GAN论文中英对照翻译--(Attentive Generative Adversarial Net ...
云检测2020：用于高分辨率遥感图像中云检测的自注意力生成对抗网络Self-Attentive Generative Adversarial Network for Cloud Detection
用于高分辨率遥感图像中云检测的自注意力生成对抗网络Self-Attentive Generative Adversarial Network for Cloud Detection in High R ...
GAN（Generative Adversarial Network，GAN）模型应用
GAN(Generative Adversarial Network,GAN)模型人脸应用目录 GAN(Generative Adversarial Network,GAN)模型人脸应用人脸重构: ...
GAN（Generative Adversarial Network，GAN）模型之：SeqGAN、IRGAN、StackGAN、BigBiGAN、GAIN模型
GAN(Generative Adversarial Network,GAN)模型之:SeqGAN.IRGAN.StackGAN.BigBiGAN.GAIN模型目录 GAN(Generative A ...
GAN（Generative Adversarial Network，GAN）模型之：EBGAN、PGGAN、CGAN、ACGAN模型
GAN(Generative Adversarial Network,GAN)模型之:EBGAN.PGGAN.CGAN.ACGAN模型 EBGAN EBGAN的思想是,将D网络看做一个能量方程.当G网 ...
GAN（Generative Adversarial Network，GAN）模型之：InfoGAN、WGAN、DCGAN模型
GAN(Generative Adversarial Network,GAN)模型之:InfoGAN.WGAN.DCGAN模型 InfoGAN 为了解决定制化生成的问题,提出了InfoGAN模型.In ...
生成对抗网络（Generative Adversarial Network，GAN）
生成对抗网络(Generative Adversarial Network,GAN) Generative:生成式模型 Adversarial:采取对抗的策略 Networks:网络(不一定是深度学习 ...
DivCo: Diverse Conditional Image Synthesis via Contrastive Generative Adversarial Network
模式崩溃是cGAN较为重要的问题,以往工作大多数都致力于关注latent code和生成图像之间的关系而没有关注通过不同latent code生成的图像之间的关系.MSGAN(mode seeking ...

A feature-supervised generative adversarial network for environmental monitoring during hazy days

A feature-supervised generative adversarial network for environmental monitoring during hazy days

A feature-supervised generative adversarial network for environmental monitoring during hazy days相关推荐

最新文章

热门文章