Looksmaxxing Forum Logo

Join 54,000+ Looksmaxxing Members!

Register a FREE account today to become a member. Once signed in, you'll be able to participate on this site by adding your own topics and posts, as well as connect with other members through your own private inbox.

Register Log in
  • DISCLAIMER: DO NOT ATTEMPT TREATMENT WITHOUT LICENCED MEDICAL CONSULTATION AND SUPERVISION

    This is a public discussion forum. The owners, staff, and users of this website ARE NOT engaged in rendering professional services to the individual reader. DO NOT use the content of this website as an alternative to personal examination and advice from licenced healthcare providers. DO NOT begin, delay, or discontinue treatments and/or exercises without licenced medical supervision. Learn more

Need participants for a survey to gather data on attractiveness

over0 said:
He's got a good eye area and jaw just has a bad nose and lips I'd pretty much agree
Click to expand...
Yeah I think the same
I think of it like this

sd = standard deviations above or below the mean
5 = 0sd (mtn)
6 = 1sd (high mtn/low htn boundary)
7 = 2sd (high htn/low cl boundary)
8 = 3sd (high cl/low chad boundary)

And just to be humorous and entertain these terms:
9 = 4sd (high chad/low giga chad)
10 = 5sd (psl god)

1sd aka 6/10 is the 84th percentile, this guy would be the most attractive looking person out of every 6.25 young (20s-30s) males

Or aka the top 4 attractive looking 20s-30s males out of every 25

That doesnt seem ludicrous as long as he dresses well and styles himself
 
sjnny404 said:
Yeah I think the same
I think of it like this

sd = standard deviations above or below the mean
5 = 0sd (mtn)
6 = 1sd (high mtn/low htn boundary)
7 = 2sd (high htn/low cl boundary)
8 = 3sd (high cl/low chad boundary)

And just to be humorous and entertain these terms:
9 = 4sd (high chad/low giga chad)
10 = 5sd (psl god)

1sd aka 6/10 is the 84th percentile, this guy would be the most attractive looking person out of every 6.25 young (20s-30s) males

Or aka the top 4 attractive looking 20s-30s males out of every 25

That doesnt seem ludicrous as long as he dresses well and styles himself
Click to expand...
that man is 17 tho
 
sjnny404 said:
Ig like late teens to young adults then i dunno

Im just tryna say between the ages in which a guy looks clearly not prepubescent nor aged ig

Idk now that you said his age you made it weird bro
Click to expand...
wait really ?
 
This ones an interesting one
4.643/10 for tom holland

I wonder if i shouldnt penalize asymmetry, but i should disproportionately reward symmetry, because i dont think hes below average and i think its due to the symmetry, or at least him being well proportionate

cd554e887ced3a7b0445f898a3e07ab0.jpg
download (7).png
 
I mean obviously it's gonna be slightly innacurate since it doesn't take onto consideration halos like colouring, lashes, certain health markers etc. Another thing is alot of these rating systems struggle rating certain ethnicities
 
sjnny404 said:
I don't need too many people to complete it, because I also plan to ask people who aren't involved in blackpill (since bpers may already be somewhat biased) or blackpill adjacent circles.

You will be shown a dataset of faces and you will give ratings to those faces. It may be around 100 faces but I'll make the UI easy to use and automatically go to the next face, etc. If you choose to participate please go on intuition only. Please do not hyperanalyze features. I will minimize things like appeal as much as possible (for example, I won't be using various ethnicities, to ensure its mainly about facial ratios).

Anyways, if anyone is interested just lmk in this thread
Click to expand...
I will
 
@Dean
Long story if you wanna read
1. I was trying to use mainly traditional data analysis methods but I realised in order to fully document every possible synergistic effect of 20+ features (including the yaw/pitch/roll of the face to account for margins of error when someone turns their head slightly or hides parts of their face with their hair) i would need to add on some kind of AI model no matter how primitive it is bc its the only way to hit every combo
2. The extra data i got is messy and not normalised which you might think is fine because AI SHOULD be able to LEARN thats the whole point of this robochimp we call the greatest advancement of mankind because why the fuck would the colour of the background of the training images affect the output accuracy of a model and introduce bias well apparently it does because AI has the brain power of a child conceived from pasteurised NUT that doesn't know the difference between jackshit and fuckall so tldr i have to normalise skin and eye colour
3. I ended up training a really primitive model that was essentially a glorified nest of yes/no switches but it was hot steaming garbage so i had to scrap it and fine tune some big opensource model's lora adapters but that meant i had borrow compute $$$ from this other guy because it was way more than just a few decision agorithms and i was dead broke (and he offered btw, i wasn't trying to swindle him or anything) but then he started being shady as hell and so i had to dip bc it got weird and as of now i have 5 google accounts i switch between since each of them has a free hour of compute per day
4. I was sick for a long time and went to the hospital thought i was gonna die etc
5. I actually FAILED my midterms (like genuinely failed, the average midterm grade for the class was 50%) and my classes aren't curved, this predictament has actually surfaced a survival instinct so old that the time for which the instinct was dead was longer than i've been alive

TLDR of the TLDR nothing beats a jet2holiday
 
Last edited:
sjnny404 said:
@Dean
Long story if you wanna read
1. I was trying to use mainly traditional data analysis methods but I realised in order to fully document every possible synergistic effect of 20+ features (including the yaw/pitch/roll of the face to account for margins of error when someone turns their head slightly or hides parts of their face with their hair) i would need to add on some kind of AI model no matter how primitive it is bc its the only way to hit every combo
2. The extra data i got is messy and not normalised which you might think is fine because AI SHOULD be able to LEARN thats the whole point of this robochimp we call the greatest advancement of mankind because why the fuck would the colour of the background of the training images affect the output accuracy of a model and introduce bias well apparently it does because AI has the brain power of a child conceived from pasteurised NUT that doesn't know the difference between jackshit and fuckall so tldr i have to normalise skin and eye colour
3. I ended up training a really primitive model that was essentially a glorified nest of yes/no switches but it was hot steaming garbage so i had to scrap it and fine tune some big opensource model's lora adapters but that meant i had borrow compute $$$ from this other guy because it was way more than just a few decision agorithms and i was dead broke (and he offered btw, i wasn't trying to swindle him or anything) but then he started being shady as hell and so i had to dip bc it got weird and as of now i have 5 google accounts i switch between since each of them has a free hour of compute per day
4. I was sick for a long time and went to the hospital thought i was gonna die etc
5. I actually FAILED my midterms (like genuinely failed, the average midterm grade for the class was 50%) and my classes aren't curved, this predictament has actually surfaced a survival instinct so old that the time for which the instinct was dead was longer than i've been alive

TLDR of the TLDR nothing beats a jet2holiday
Click to expand...
when do you think it will be finished?
good luck on ur midterms bro that sucks
 
blanquita444 said:
when do you think it will be finished?
good luck on ur midterms bro that sucks
Click to expand...
Honestly no clue, i just spent the last 24 hrs on this shit and lost my mind

TLDR appeal is law

I got a bigger database with a bunch of randoms and not just slightly altered versions of the same 5 people

I did two versions of an algo

One with only ratios and the other includes pheno (eye colour) etc

Ratios accounted for only 33% of the variance of scores for looks but ratios with pheno is 80%+
 
I realised also that i shouldnt negatively weigh things like angled/weirdly positioned face, because that would make me lose more accuracy. Sine angles generally make someone look better

So what I think I have to do is, basically angles should increase scores technically, but also introduce a higher margin of error on how accurate that could be

But then to find the actual margin of error I'm pretty sure I have to find a database of people who took a picture and looked straight into the camera, and then a few other pictures where they angled their face to the side (in increments), and then were rated by others, then I could maybe get a quick correl on what degree of angles could introduce bias, and then I'll be able to create a data backed error rate

But who the fuck has ever made a study on that

So for now I do have face angle as a feature, and threw in some random number on how much it can affect a rating

Obviously angled faces will fuck up ratios but i had to make a compromise bc

a. I dont have enough data to filter out every slightly angled photo i have I'd be left with maybe 30 data points max
b. Most people who send selfies for rate tend to angle their face somewhat anyways, so i might as well put that as a factor in someones score, just increase error rate of the ratings for it

Anything that will hide ratios like hair will increase error so having hair kinda covering the face is also a feature i added

I havent slept i dont know how much sense im making
 
download (9).png
Example of
Score including pheno = 5.8
Score with only ratios = 5.6
 
You must log in or register to reply here.

Users who are viewing this thread

Total: 1 (Looksmaxxers: 0, Guests: 1)
Back
Top