Hats off to whoever took time to create this (and the beautiful whippet!).

[koneko@toot.beep.computer]

@Natasha_Jay

[un_bourguignon@piaille.fr]

@Natasha_Jay
Ping @emeline
@Milena_Hime

[emeline@piaille.fr]

@un_bourguignon Merci pour le ping, je crois que j'en avais vu passer un similaire. J'adore !

@Natasha_Jay @Milena_Hime

[un_bourguignon@piaille.fr]

@emeline
J'ai trouvé ça très drôle et bien vu en plus .
@Natasha_Jay @Milena_Hime

[pmiossec@piaille.fr]

@pork_soda @srtcd424 We say "Chocolatine" and not "Pain au chocolat" #ExportFrenchDebate 🤪

[tomulne@framapiaf.org]

@Natasha_Jay
For me it's like :

[alice@lgbtqia.space]

@agowa338 these used to be used as adversarial training sets for computer vision algorithms (not this one, probably, but Chihuahua or blueberry muffin, Sharpei or towel, Labradoodle or fried chicken, etc). There are whole image sets of them.

@Natasha_Jay

[alice@lgbtqia.space]

@Natasha_Jay I love these adversarial image sets

@darkphoenix

[agowa338@chaos.social]

@alice @Natasha_Jay

And captures are often used to label such datasets.

E.g. You know the result to 80% of the shown images and let the user click all of them. The user doesn't know which ones you expect them to click. Then for the few unlabeled ones you record if they did/didn't click them and then you serve the same unknown ones to multiple people and bam you've a very good way to enlarge your datasets.

That's how google originally started with recaptures and what they're doing now...

[alice@lgbtqia.space]

@agowa338 yup, crowd-sourced ML training.

@Natasha_Jay

[stephaniepixie@fandom.garden]

@alice Sloth au chocolat

[isaacfreeman@cloudisland.nz]

@cainmark Yeah, we all came here to post this. Whippet indeed good. Good whippet.
@phil_stevens @Natasha_Jay

[bot_anix@nrw.social]

@Natasha_Jay gut kombiniert ️

[bot_anix@nrw.social]

@pork_soda @srtcd424 me, sometimes