My understanding is that the 'truncation' parameter makes the output of BigGAN more closely resemble the training data set. Moving the truncation towards 0 wil

My understanding is that the 'truncation' parameter makes the output of BigGAN more closely resemble the training data set. Moving the truncation towards 0 will give you more realistic but less varied images, while moving it towards 1 will give you more varied but increasingly potentially nonsensical images.