Why do computer generated faces differ from real ones?

[kolleamm]

I've probably been wondering about this questions for years now. Why is it that even some of the best works fall short of looking undoubtedly real?

Whether the face is plastic, or just on a screen, why is it that it looks unreal? Is it the lack of fine detail?

I know there are exceptions but let's consider just the majority.

 

[jedishrfu]

This is known as the uncanny valley, when you think something that looks very human also looks or behaves in a creepy and unnerving way:

https://en.wikipedia.org/wiki/Uncanny_valley

 

[kolleamm]

That's true. Just looking at a photograph though, what would make it seem off though? The first image has a good amount of detail and good shading, so what would be missing?

 

[rollete]

Hardware is a limitation in real-time rendering applications. Consoles and PCs still lack the horse power to render truly life-like CGI on the spot. There's also the fact that faces are particularly tricky objects to model and animate realistically.

The best CGI is found in pre-rendered content--any type of static or motion picture.

Who's real, who's CGI?

 

[Bystander]

Who's real, who's CGI?

Just a stab in the dark ... first (male) is CGI, second (female) is real.

 

[OmCheeto]

This is known as the uncanny valley, when you think something that looks very human also looks or behaves in a creepy and unnerving way:

https://en.wikipedia.org/wiki/Uncanny_valley

Sounds similar to the hatred of "Mimes".

Though, professional mimes claim that they never really met someone that hated them.

Mime here.
I've never met a person (while I was in mime) that seemed to not like mimes.

[ref: https://www.reddit.com/r/OutOfTheLoop/comments/284ux6/why_do_many_people_hate_mimes/]But, my theory is, that the people they met, were just being polite.
Machines should look and act like machines, and people should look and act like people.

Everything else, is just creepy.

ps. See: Clowns. What the hell are clowns supposed to be, anyways.

[edit] Ha! What did I tell you...

Which do you hate more: Mimes or Clowns?

• Mimes
• Clowns
• I'm an Equal Opportunity Hater

[ref: http://www.ihatemimes.com/]​

 

[davenn]

That's true. Just looking at a photograph though, what would make it seem off though? The first image has a good amount of detail and good shading, so what would be missing?

if that second photo ( the woman) is real and not a manikin ( I almost suspect it is, unless you can show me otherwise)
then the photo editor has just gone overboard in photoshop ... nothing to do with the way the screen is rendering the image

there is hardly a glamour/fashion image that isn't photoshopped these days before displayDave

 

[kolleamm]

The woman in the second photo is not real, it's silicone skin :D

 

[rollete]

Just a stab in the dark ... first (male) is CGI, second (female) is real.

Both CGI. I'd peg the woman real too if I didn't know it's CGI.

http://hexus.net/tech/news/graphics/53617-activisions-realistic-3d-face-rendering-technology-video/

 

[russ_watters]

That's true. Just looking at a photograph though, what would make it seem off though? The first image has a good amount of detail and good shading, so what would be missing?

I disagree. Both of those photos are lacking in sufficient detail. But more importantly, the facial expressions are off, which is is the giveaway for me.

Both in #4 look real to me. The lack of expression probably helps, but the texture complexity makes them far superior to the pictures in the OP.

 

[Bystander]

Both CGI.

The male has no "frown lines;" that's a "gimme." From the way the question/challenge was constructed, I had to gag down objections to the female (ears) and pick it as "live."

 

[jedishrfu]

The first woman in the OP post is actually an android likeness of a popular Japanese announcer shown here along with the roboticist creator:

http://www.dailymail.co.uk/sciencet...apanese-scientists-unveil-female-android.html

The First woman is not CGI and not alive either.

 

[Greg Bernhardt]

Why is it that even some of the best works fall short of looking undoubtedly real?

The first graphic looks like it's from Dead Space which is like 8 years old now and optimized so many systems can run it. Some of the top of the line look pretty darn scary close. Give it another 10 years and you won't know the difference.

https://www.fastcodesign.com/3047730/this-is-the-most-realistic-computer-generated-skin-weve-ever-seen

 

[DaveC426913]

More than virtally anything else, humans are adept at examining faces. A large part of our processing power is devoted to this.

Ther are 7 billion people in the world, and yet we can identify a friend's face at 20 yards from virtually any angle except directly behind him, in a fraction of a second.

It's what we do.

 

[davenn]

The woman in the second photo is not real, it's silicone skin :D

as I suspected

so why are you trying to compare two totally different situations and then blame the computer for not giving a realistic outcome

a little pointlessDave

 

[jedishrfu]

I suspect that Oliver Sacks wouldn't recognize the difference as he was face blind.

 

[OCR]

I suspect that Oliver Sacks wouldn't recognize the difference as he was face blind.

Interesting...[COLOR=#black]..[/COLOR] The[/PLAIN] Man Who Mistook His Wife for a Hat.

Paul Dirac suffered from the same disorder...
https://en.wikipedia.org/wiki/Prosopagnosia

 

[Psinter]

The short answer to the title question is: Mainly because of the rendering engine used and how it is programmed.

However, it also depends on the 3D graphics API (see a list here) used by the renderer and like mentioned above, how that library is used by the programmer/s. Another big dependency as to why they differ from real ones is that a lot of it will be decided on how the programmer or graphic designer works with the software (3D computer graphics software). That software can be a game engine, a 3D modeling software, a CAD software, or an exclusive 3D rendering software.

The reason it is this way is because someone programs an application that communicates with a rendering library, that rendering library communicates with a driver and then that driver communicates with the GPU (Graphics Processing Unit). This is a very lengthy subject. If you are interested, the keywords you are looking for is: Rendering Pipeline. Here is a very basic slide in which you could maybe find more keywords to search and study: https://www.ics.uci.edu/~majumder/VC/classes/geomgraphics.pdf

The subject itself is complex. Not for being lazy, but I can tell you that if you read a book of Mathematics for Game Developers you will see part of the whole mathematic world behind it as well as the plethora of ways to approach 3D rendering. It gets complex pretty fast and some things you simply cannot understand unless you are well versed in differential equations (referring specifically to animations and real time rendering). For example, one filtering in texture mapping make the rendering hardware examine some partial derivatives of interpolated components of a texture and it does so with a given differential equation. Yet this is only one method of texture mapping. It would be the programmer/s of the rendering engine, the users of the rendering engine, or the designer/s, the ones to choose which method to use for mapping and that will have different results in the final rendered image.

And so on goes the list of variables that affect the rendering of objects.

All in all, the answer is that there are many variables that affect the rendering.

If you ask for my opinion, which you did not, I would tell you that based on my understanding and knowledge, we have the mathematics and tools required to render resembling enough faces to a level in which they cannot be differentiated. What we don't have for the public (you and me) is the widespread access to the work of individuals with intellect enough to use that mathematic and tools to create or help create something that astonishes us and makes us think it is real. Rendering something like that would require a huge human effort and intellect; and things that require huge human efforts and intellect are scarce.

Hardware is a limitation in real-time rendering applications. Consoles and PCs still lack the horse power to render truly life-like CGI on the spot. There's also the fact that faces are particularly tricky objects to model and animate realistically.

I beg to partially differ on the first sentence. I think the mathematics and tools are more than good enough. What is not good enough is the effort put into making the software and the effort of the designer of the models like I mentioned above. On the second sentence I agree, real time rendering of highly realistic objects with and ordinary user computer at the moment of writing this is very hard because it lacks the resources.

In your third sentence I also agree, that is a fact and variable to take into consideration when looking at OP's question. It is tricky and like I mentioned above, requires huge human efforts which are scarce.

The best CGI is found in pre-rendered content--any type of static or motion picture.

True. Pre-rendered content contains more detail that would be hard or would require more processing resources than an ordinary individual's computer would have available for real time rendering. Hence, looking better to the human eye. That's why some game engines give you the option to pre-render content and pack it in the game which increases the size of the game, but gives more realistic graphics.

Who's real, who's CGI?

I think both of them are CGI. The first one has a logo, the second one's scene was not rendered real enough for my eyes and the lightning shades of the neck are not on par with the amount of light striking the lower eye and nose.

Had the first one not had a logo I could have mistaken it by real. Yet a closer inspection shows lightning in the neck muscles that give away the mesh construction.

 

[Pepper Mint]

I think he is real but she isn't. Look at her ear which is graphically constructed. Its border needs to be blurred out a little more to make her look more real.
On the other hand, the man was naturally constructed with asymmetrical ears. He doesn't look handsome with wrinkles and there seem to be two light sources on him (from the back and the top) though.

 

[kolleamm]

Looking at the animation in the link in #13, it does make you realize what a big role the stretch marks in our skin play in convincing us a face is real. I do realize my second picture in the first post was not CGI, but still worth mentioning nevertheless. If you look at the lips they have virtually no stretch marks or ridges.

 

[kolleamm]

My ultimate goal is to be able to print out a 3d face model in plastic and make it look realistic, whether flexible or not it doesn't matter. Therefore I thought the best place to start would be to ask why most CGI is so easily distinguishable from the real.

What now appears to be evident to me is that lighting and a lack of detail plays a big role between real and unreal.
For example the way that light reflects off our moist lips, and the tiny stretch marks on our skin (which a computer will have a hard time generating with correct shading and not mere lines)

Here is a good comparison

 

[Greg Bernhardt]

Here is a good comparison

True, but your CGI example on the left is really poorly done. There are CGI examples out there that make you look twice.

 

[kolleamm]

It's amazing how real they look yet at the same time my mind is able to know they aren't. Lack of detail again perhaps? Or lighting?

 

[Greg Bernhardt]

It's amazing how real they look yet at the same time my mind is able to know they aren't. Lack of detail again perhaps? Or lighting?

Yeah we haven't gotten there yet, but it won't be long. I think the middle one of Kristen Stewart is about as good as I've seen. If I didn't see that image in a thread about CGI then I may think it was real.

 

[axiom]

From my limited experience with 3D applications the main reason faces are different is the lighting, and how the lighting is reflected [specularity]. Also the the skins texture and mapping, 3D skin can sometimes appear flat and too sharply defined whereas real skin or a quality 3d rendering the subject has a certain depth that the eye recognises...

 

[cosmik debris]

I think also facial symmetry is a giveaway unless it is purposely given attention.