图像评价常用指标(PSNR、SSIM、LPIPS 、FID、Recall)

最近需要测试图像的指标,然后特意去关注了一下相关的信息,然后主要用的评价指标为:

1、PSNR 和 SSIM

去看:图像质量评价指标: PSNR 和 SSIM_马鹏森的博客-CSDN博客_psnr范围

5、LPIPS 

也叫作感知损失,比PSNR和SSIM更接近于真实人感到的感觉:

https://blog.csdn.net/weixin_43135178/article/details/127664187

3、FID

去看:FID使用(Frechet Inception Distance score)_马鹏森的博客-CSDN博客

出自论文:GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium(https://arxiv.org/abs/1706.08500)

4、Recall(多样性)

代码:https://github.com/blandocs/improved-precision-and-recall-metric-pytorch

关于这个评价参数我是在“Diffusion-GAN: Training GANs with Diffusion”论文中看到的,地址为:https://arxiv.org/abs/2206.02262

原文对这个的解释:Lower FIDs indicate better fidelity, while higher Recalls indicate better diversity. We further report the improved Recall score introduced by Kynkäänniemi et al. [2019] to measure the sample diversity of generative models.

出自论文Improved precision and recall metric for assessing generative models(https://arxiv.org/abs/1904.06991

注意:这个只能用于“Evaluation of realism score using StyleGAN and FFHQ dataset can be run with:”。似乎对其他的数据集没有用,我刚试了一下其他的数据集,不行!

你可能感兴趣的:(机器学习基础,图像转换,1024程序员节,深度学习,python)