Best Setup

[jS5t3r]

Hello,

Thank you for sharing your code. I am thinking of making a comparison with my own method.

Do you think that on the FFHQ dataset, this is a fair setup?

"latent_size": 128,
"lr": 0.0001,
"kd_measure": "kd",
"kd_weight": 1.0,
"kd_lambda": 1.0,
"epochs": 250,
"early_stop_patience": 35,
"early_stop_min_epochs": 20,
"batch_size": 32,
"effective_batch_size": 32,
"identity_latents": "/projc/mad/idistill_latents128/ffhq_latents.npz"

Thanks in advance! :)