What's new

[Forum Event] Draw a Star

Which one? And don't say voting, as it's not always objective but may depend on how many friends you'll gather to vote for you.

Should just be whatever looks most like the person given, so in the first case, Scarlett Johansson, relative to the reference used. Which yeah, is subjective to you the person running this, but I mean, you can't expect it not to be when you have one person running an event.
 
IGN: Zackario

JustineArcherage.jpg
 
Which one? And don't say voting, as it's not always objective but may depend on how many friends you'll gather to vote for you.
Which one? yea, that's what I'd want to know as well which might work in this case. Voting obviously won't work.

Ok, this is the only potential way I found which allows comparing two faces:
Betaface Demo
GPglLOY.png


I hadn't heard of this org before so I decided to check on its accuracy. I compared all submitted pics of the star of the day. Here are the results, which shows a high similarity of these faces.
Results for Scarlet:
MsG95jr.png

Results for Justin:
pTjL9f9.jpg


It has some famous companies as its customers (including 20th Century Fox & Disney) which uses their technology. Based on this and the results mentioned above I'd say that their face recognition tech is sufficiently accurate.

With that out of the way, I compared each reference star image with the corresponding character image. Apologies if I missed some, just took few samples to experiment with.
FuMqBRI.jpg

EVATSpy.jpg


Heres what I propose:
- Every participant can make their character and use that link to get highest similarity percentage that they can and feel comfortable with.
- They upload their final character and reference images as they currently do.
- A participant may feel the need to improve their submission after someone else's submission with a better percentage. You can allow/restrict re-edits, upto you.
- In case of a tie, @Sparkle can take the final decision

Extra notes:
- Ikr it wrongly classifies age consistently, but that isn't our concern and is more difficult to predict on stars who strive for a youthful look.
- If you note the case for Day-1, @Bell's submission has the highest percentage which is strange. But on overlaying the two images it was clear that their software disregards skin complexion to a certain degree, so that's a minor drawback in my opinion.

Here is the overlay gif and its vid link in case gif doesn't work
Q1A8Le2.gif

http://i.imgur.com/1nHThYk.mp4
 
Last edited:
Which one? yea, that's what I'd want to know as well which might work in this case. Voting obviously won't work.

Ok, this is the only potential way I found which allows comparing two faces:
Betaface Demo
GPglLOY.png


I hadn't heard of this org before so I decided to check on its accuracy. I compared all submitted pics of the star of the day. Here are the results, which shows a high similarity of these faces.
Results for Scarlet:
MsG95jr.png

Results for Justin:
pTjL9f9.jpg


It has some famous companies as its customers (including 20th Century Fox & Disney) which uses their technology. Based on this and the results mentioned above I'd say that their face recognition tech is sufficiently accurate.

With that out of the way, I compared each reference star image with the corresponding character image. Apologies if I missed some, just took few samples to experiment with.
FuMqBRI.jpg

EVATSpy.jpg


Heres what I propose:
- Every participant can make their character and use that link to get highest similarity percentage that they can and feel comfortable with.
- They upload their final character and reference images as they currently do.
- A participant may feel the need to improve their submission after someone else's submission with a better percentage. You can allow/restrict re-edits, upto you.
- In case of a tie, @Sparkle can take the final decision

Extra notes:
- Ikr it wrongly classifies age consistently, but that isn't our concern and is more difficult to predict on stars who strive for a youthful look.
- If you note the case for Day-1, @Bell's submission has the highest percentage which is strange. But on overlaying the two images it was clear that their software disregards skin complexion to a certain degree, so that's a minor drawback in my opinion.

Here is the overlay gif and its vid link in case gif doesn't work
Q1A8Le2.gif

http://i.imgur.com/1nHThYk.mp4

I was going to propose something like this too but didn't have the time to look into it. I think it's a great idea!
 
"If you note the case for Day-1, @Bell's submission has the highest percentage which is strange."

K. Regardless of the fact that, no, actually, I was 4.8% lower than the highest, howabout not calling people out and being like "Huh that's weird that it thinks that's so close cause it's clearly not" which is basically what you just said.

Either way, pretty sure this is just a "Mod's Choice" so I wouldn't push for too much "Super specific official judging". You're not even taking part in this event so just let people have fun without turning this into a "Try to get the highest score on a computer generated score machine" challenge.
 
Last edited:
' If you note the case for Day-1, @Bell's submission has the highest percentage which is strange. But on overlaying the two images it was clear that their software disregards skin complexion to a certain degree, so that's a minor drawback in my opinion.'
This just tells you the software isn't that accurate and does not give a proper result on how close the 2 images are so it should not be used.
 
it's a good idea but it's too technical, where's the fun in that? the whole event should be all about fun... loosen up sometimes sir :);)
 
"If you note the case for Day-1, @Bell's submission has the highest percentage which is strange."

K. Regardless of the fact that, no, actually, I was 4.8% lower than the highest, howabout not calling people out and being like "Huh that's weird that it thinks that's so close cause it's clearly not" which is basically what you just said.
Ok, indeed I could've used a better choice of words for that and I didn't know that mentioning someone can call out (or alert) them in any way. Sorry about that. My intention for mentioning your character in specific was to actually support that your char looks similar, but can seem otherwise at a glance. I made that gif to show that your char was real close in terms of facial features.

Either way, pretty sure this is just a "Mod's Choice" so I wouldn't push for too much "Super specific official judging". You're not even taking part in this event so just let people have fun without turning this into a "Try to get the highest score on a computer generated score machine" challenge.
Sparkle asked for a suggestion, so I gave one. I'd certainly take part but a tedious task to get close to how a real star looks, maybe I'd give it a try if I really like the star of the day enough to be motivated into creating his/her char avatar.
Well, people are comfortable with varying degrees of specificity and quantification of the stuff they do, so maybe it seems that I'm ruining the fun for you by suggesting a score system while some may think that its a great idea. I won't say anything to that since 'fun' is subjective.
 
' If you note the case for Day-1, @Bell's submission has the highest percentage which is strange. But on overlaying the two images it was clear that their software disregards skin complexion to a certain degree, so that's a minor drawback in my opinion.'
This just tells you the software isn't that accurate and does not give a proper result on how close the 2 images are so it should not be used.
I mentioned that point specifically to say that, yes, it might be inaccurate in terms of skin complexion, but then presented that gif to show that two images were quite close in other facial features
 
perhaps you should stop taking things so seriously and just have some fun..
it's a good idea but it's too technical, where's the fun in that? the whole event should be all about fun... loosen up sometimes sir :);)
Yea, frankly I wasn't planning on getting too technical. But if I had just given the link and suggested to use that, it'd be an ingenuine suggestion. Moreover it kinda got interesting along the way and before I knew it, I had gone the extra mile XD
 
@Necrote thank you for the research. We'll not be changing the event that is already going but may take your idea in future.
 
Back
Top