PULSE

motivation

• LR-HR image pair
• MSE loss leads to smooth result
  “ As a result, MSE should not be used alone as a measure of image quality for super-resolution”
• “the lack of detail in algorithms that rely only on an lp norm cannot be fixed simply by changing the architecture of the network”
• main idea: use unconditioned GAN to generate SR instead of using conditioned GAN to add details to transform LR to SR

method

传统方法通过减少L1/L2范式来拉近SR与HR之间距离，但距离缩小并非意味着属于真实图像域，论文方法通过讲真实域上图像逐渐拉近HR来生成更具有细节表现的SR

通过使用styleGAN(experiment)来生成真实域上图像，随后通过优化Latent Space来使其DS后与目标LR相似

DS loss

Latent space

experiment

Results

Robustness

conclusion

• 使用unconditioned GAN解决SR问题
• 通过真实图片域中优化latent space来逼近HR
• 实现大倍数超分辨率问题（x64）