FID
真实图与生成图的两分布间的距离
pip install pytorch-fid
python -m pytorch_fid path/to/dataset1 path/to/dataset2
把生成图片的路径和真实图片的路径放进去即可,和顺序无关。
也可以选择与–dims N标志一起使用的特征维数,其中N是特征的维数。
python -m pytorch_fid path/to/dataset1 path/to/dataset2 --dims 2048
LPIPS
学习感知图像块相似性,用来测量生成图像的多样性
pip install lpips
import lpips
loss_fn_alex = lpips.LPIPS(net='alex') # best forward scores
loss_fn_vgg = lpips.LPIPS(net='vgg') # closer to "traditional" perceptual loss, when used for optimizationimport torch
img0 = torch.zeros(1,3,64,64) # image should be RGB, IMPORTANT: normalized to [-1,1]
img1 = torch.zeros(1,3,64,64)
d = loss_fn_alex(img0, img1)
参考: