3 ms·
I assume the risk of giving the possibility of generating "person of group X doing something bad" ? Or more prosaically, we're MUCH better at noticing somethin
by BlueTemplar 5y ago
I assume the risk of giving the possibility of generating "person of group X doing something bad" ?
Or more prosaically, we're MUCH better at noticing something wrong in humans (especially faces !) than in, say, puppies, so that would make DALL-E 2 look bad to an uninformed observer.
- gwern 5y agoIt's the PR risk. You can easily pass the 'look good to an uniformed observer' test with human faces. Remember, faces were something GANs were doing near-flawlessly back as far as ProGAN all the way back in the dark ages of late 2017. (Then StyleGAN did non-photograph faces, like anime - see my ThisWaifuDoesNotExist for a demo of that.) Doing them as part of a larger composition, where the face is a small part of the images and may vary much more than in closeup centered portraits, is harder, but take a look at how Facebook's DALL-E rival Make-A-Scene does it: https://arxiv.org/abs/2203.13131#facebook https://arxiv.org/abs/2203.13131#facebook They specially target faces as part of the training process, with face-specific detectors/losses, and so the faces come out great.
- BlueTemplar 5y agoFair enough, but it could be... both ! Since you know that training for photo-realistic faces is going to take extra effort, and that you're not going to allow them - you could just spare that effort ! (Or maybe leave it for later, if the network is flexible enough ?)