|
|
|
|
|
|
by bgie
2843 days ago
|
|
|
Not my field of expertise but on page 15 they claim " In our reconstruction analysis, we used a pre-trained DGN which was provided by Dosovitskiy & Brox"
So if the deep dream generator is not trained on their actual (limited) training data but instead on a massive dataset of 'all photos that are real", I would say the study is valid. |
|
|