About the experiment

[zy-charon]

Hi and thanks for the great work. I have two questions about the experiment. 1. What is the division ratio of Sin and Sreal when calculating FID? Whether the three datasets randomly select 30 reference images for each unseen category, and then generate 128 images. 2. For the experimental part "Augmentation for Classification" , is the generated image using the image in Dtrain as reference images?

[edward3862]

Hi, sorry for the late reply.

1. The ratio is 0.5 as specified in the config file. Yes, you can find the code in main_metric.py.
2. Yes, the source images to do augmentation is from the training set.