Software Engineering Courses (SECourses)•3y ago

Hello! New here. Dreambooth training question. If I'm trying to train a model on images of a woman w

Hello! New here. Dreambooth training question. If I'm trying to train a model on images of a woman who doesn't have a thin body type, is having regularization images of thin women influencing the training towards a thin body type? I've been banging my head on this training for 3 days now with no real success.

AIGambino•12/26/23, 10:11 PM

Disabled the AV.

AIGambino•12/26/23, 10:27 PM

well this is as far as I can get it hangs here

ZZax Hello! New here. Dreambooth training question. If I'm trying to train a model on...

Furkan Gözükara SECourses•12/26/23, 10:43 PM

have you tried our reg images?

Furkan Gözükara SECourses•12/26/23, 10:43 PM

it has mixed body types

AAIGambino well this is as far as I can get it hangs here

Furkan Gözükara SECourses•12/26/23, 10:43 PM

good working

FFurkan Gözükara SECourses have you tried our reg images?

ZaxOP•12/26/23, 10:50 PM

hm, I will take a look at your reg images, I'd been using Aitrepreneur's images but it looks like your dataset is much larger (5200 images vs 1000 for his)

ZaxOP•12/26/23, 10:51 PM

also, do you have tips for number of images vs repeats? Is there some target I'm aiming for like images*repeats=500, or something like that?

ZaxOP•12/26/23, 10:53 PM

oh yeah these images are way more diverse. Ethnicities, ages, body shapes. I'll have a try with these. Thanks!

ZZax hm, I will take a look at your reg images, I'd been using Aitrepreneur's images ...

Furkan Gözükara SECourses•12/26/23, 10:53 PM

my images are manually collected

Furkan Gözükara SECourses•12/26/23, 10:53 PM

i spent over 5 days

ZZax also, do you have tips for number of images vs repeats? Is there some target I'm...

Furkan Gözükara SECourses•12/26/23, 10:53 PM

i suggest repeat 200

Furkan Gözükara SECourses•12/26/23, 10:54 PM

and save based number of steps

Furkan Gözükara SECourses•12/26/23, 10:54 PM

so 1 epoch

Furkan Gözükara SECourses•12/26/23, 10:54 PM

200 repeat

ZaxOP•12/26/23, 10:54 PM

repeat 200 regardless of the number of images?

ZZax repeat 200 regardless of the number of images?

Furkan Gözükara SECourses•12/26/23, 10:54 PM

how many train images you have?

ZaxOP•12/26/23, 10:54 PM

Furkan Gözükara SECourses•12/26/23, 10:54 PM

ok it is quite a lot

Furkan Gözükara SECourses•12/26/23, 10:55 PM

make repeat 130

Furkan Gözükara SECourses•12/26/23, 10:55 PM

because otherwise you will need more reg images

Furkan Gözükara SECourses•12/26/23, 10:55 PM

also 130 epoch is decent for 39 images

ZaxOP•12/26/23, 10:55 PM

is there an ideal ratio of reg images to training images?

ZZax is there an ideal ratio of reg images to training images?

Furkan Gözükara SECourses•12/26/23, 10:55 PM

actually the more you use it is better

Furkan Gözükara SECourses•12/26/23, 10:55 PM

therefore we try to use max

Furkan Gözükara SECourses•12/26/23, 10:55 PM

but as you do more repeat

Furkan Gözükara SECourses•12/26/23, 10:55 PM

it will do more epochs

Furkan Gözükara SECourses•12/26/23, 10:55 PM

200 repeat actually means 200 epoch

Furkan Gözükara SECourses•12/26/23, 10:56 PM

not exactly due to scripts random image training logic but almost equal

ZaxOP•12/26/23, 10:57 PM

ok so 130*39=5070, which is close to 5200 which is the number of regularization images. Is that number similarity a coincidence or is that what I'm aiming for, for the number of repeats multiplied by the number of images to be about equal to the number of reg images?