2 ms·

Also: https://www.reddit.com/r/MachineLearning/comments/ch0qms/p_decomposing_latent_space_to_generate_custom/ https://www.reddit.com/r/MachineLearning/comments/

by gwern 7y ago

Also: https://www.reddit.com/r/MachineLearning/comments/ch0qms/p_decomposing_latent_space_to_generate_custom/ https://www.reddit.com/r/MachineLearning/comments/ch0qms/p_d...

hanniabu 7y ago
Looks like you're the creator so as some feedback it'd be nice to see a simple outline of the steps with "inactive" styling, the step you're currently on as bold (maybe with an arrow in from of it?), and as you go through it the previous steps have a checkmark appear next to it (and possibly go back to "inactive" styling). The steps could be something like this: (1) Select character, (2) Select colors, (3) Select outfit, (4) Select pose The reason I suggest this is I was a little confused on what the options were, what my future options would be, and how many options are left.
- gwern 7y ago
  To clarify, I'm not the creator. Sizigi Studios uses a dataset I put together (https://www.gwern.net/Danbooru2018 https://www.gwern.net/Danbooru2018) as their primary training corpus, AFAIK, and they were partially inspired by my application of GANs to anime art (as demonstrated on https://www.thiswaifudoesnotexist.net/ https://www.thiswaifudoesnotexist.net/ and see https://www.gwern.net/Faces https://www.gwern.net/Faces for much more detail about every aspect of it), but I have never been involved with them and don't know much about what they've done other than what you can read in OP. It tickles me pink to see people following up on my anime GANs, though, especially as a startup! The Great Work goes on.
  - kevinfrans 7y ago
    One of the creators here -- gathering the Danbooru data in one place was definitely a big help! Anime's been a pretty nice space (both personally and research-wise) especially due to the abundance of data online, and it's great that we no longer have to manually scrape image hosts (which I've spent many hours doing in the past @ https://github.com/kvfrans/deepcolor https://github.com/kvfrans/deepcolor, https://canvasdrawer.autodeskresearch.com/ https://canvasdrawer.autodeskresearch.com/ etc).
- Twisol 7y ago
  Gwern created a similar thing, but I believe it was a different team who made the Waifu Vending Machine. Not least because there are pictures, and Gwern is quite known for their pseudonymity.

by gwern 7y ago

Also: https://www.reddit.com/r/MachineLearning/comments/ch0qms/p_decomposing_latent_space_to_generate_custom/ https://www.reddit.com/r/MachineLearning/comments/ch0qms/p_d...

hanniabu 7y ago

Looks like you're the creator so as some feedback it'd be nice to see a simple outline of the steps with "inactive" styling, the step you're currently on as bold (maybe with an arrow in from of it?), and as you go through it the previous steps have a checkmark appear next to it (and possibly go back to "inactive" styling). The steps could be something like this: (1) Select character, (2) Select colors, (3) Select outfit, (4) Select pose The reason I suggest this is I was a little confused on what the options were, what my future options would be, and how many options are left.

gwern 7y ago
To clarify, I'm not the creator. Sizigi Studios uses a dataset I put together (https://www.gwern.net/Danbooru2018 https://www.gwern.net/Danbooru2018) as their primary training corpus, AFAIK, and they were partially inspired by my application of GANs to anime art (as demonstrated on https://www.thiswaifudoesnotexist.net/ https://www.thiswaifudoesnotexist.net/ and see https://www.gwern.net/Faces https://www.gwern.net/Faces for much more detail about every aspect of it), but I have never been involved with them and don't know much about what they've done other than what you can read in OP. It tickles me pink to see people following up on my anime GANs, though, especially as a startup! The Great Work goes on.
- kevinfrans 7y ago
  One of the creators here -- gathering the Danbooru data in one place was definitely a big help! Anime's been a pretty nice space (both personally and research-wise) especially due to the abundance of data online, and it's great that we no longer have to manually scrape image hosts (which I've spent many hours doing in the past @ https://github.com/kvfrans/deepcolor https://github.com/kvfrans/deepcolor, https://canvasdrawer.autodeskresearch.com/ https://canvasdrawer.autodeskresearch.com/ etc).
Twisol 7y ago
Gwern created a similar thing, but I believe it was a different team who made the Waifu Vending Machine. Not least because there are pictures, and Gwern is quite known for their pseudonymity.