>>60486535>>60486621How much stuff gets burnt in honestly just depends on the variety of the dataset rather than the tagging (I mean it helps but there is a limit) lower the amount of pictures the more inevitable of something getting burnt in.
Best way to do it is use Dadaptation training and train in epochs rather than repeats and just test each epoch until you get that's the least troublesome/most accurate.
In terms of pruning tags, people have different opinions about it but I say remove color tags as they are just useless and I've never seen color get burnt in.
>>60486894Fucking BADASS. She is about to take off her pain inhibitors.
