search

taki0112 / UGATIT

Official Tensorflow implementation of U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation (ICLR 2020)
MIT License
6.17k stars 1.04k forks source link

Can you release KID evaluation code you used in your paper? #64

Open Ha0Tang opened 5 years ago

FantasyJXF commented 5 years ago

@Ha0Tang https://github.com/taki0112/GAN_Metrics-Tensorflow The other repo of the author might help

FantasyJXF commented 5 years ago

@taki0112 Hi, taki. I use your pretrained 100epoch model to run the test with your dataset, and calculate the KID with THE above method of yours. I put 100 selfie image of testA into folder real_source, and 100 anime image from trainB into folder real_target, the generated 100 anime images from testA are put into folder fake. It takes very long time to calculate FID. 

KID_mean :  2.721179649233818
KID_stddev :  0.5364757962524891

And

mean_FID :  1.3425915359497071
mean_KID_mean :  11.149919640272856
mean_KID_stddev :  0.4221153156831861
FantasyJXF commented 5 years ago

And my question is how can I evaluate a single image?