search

cyclomon / UNSB

Official Repository of "Unpaired Image-to-Image Translation via Neural Schrödinger Bridge" (ICLR 2024)
MIT License
169 stars 8 forks source link

Question about FID on horse2zebra dataset #9

Closed liangbingzhao closed 1 year ago

liangbingzhao commented 1 year ago

I have followed your default setting to train the horse2zebra dataset, and test on the epoch 400's model. Generate 120 images(which is the same num as the test dataset) and use NFE=5 to calculate FID, only got 231.22809380899474. Is it possible for u to share how to get FID 35.7

liangbingzhao commented 1 year ago

sry, not 231, but 66.99644191929943, which is still too larger than yours

cyclomon commented 1 year ago

Hi, thank you for your interest on our research.

In Image-to-image translation tasks (especially for dataset published in CycleGAN),

the paper reports best FID score during training, not the value of epoch 400.

Therefore we followed the experimental protocol of previous works.

Please refer to the following link:

https://github.com/taesungp/contrastive-unpaired-translation/issues/65

https://github.com/lyndonzheng/F-LSeSim/issues/1