3 ms·
I think that the idea is that you use transfer learning to bootstrap from a general gan to a gan for your domain - and then use that to generate odd data.
by sgt101 5y ago
I think that the idea is that you use transfer learning to bootstrap from a general gan to a gan for your domain - and then use that to generate odd data.
- johndough 5y agoLast I checked, transfer learning for StyleGAN2 still produced somewhat wonky results if the datasets were not extremely close. Have there been improvements in this field?
- sgt101 5y agoI don't know - but I guess that must be in the paper. I had intended to read this properly after I looked at it when the story came out, but I haven't had time, sorry.