why don't computer generated graphics look real

[hahher]

in animated movies or clips, you would never be fooled into thinking that the models were real people. they always look like plastic or dolls.

since movies are pre-rendered they can basically use as much time as they want to achieve best graphics (for say just a 1 minute clip to demo potential). yet still i've never seen a convincing animation.

i remember when i first saw screenshot of final fantasy movie, with girl looking into microscope, i thought it looked real. but seeing that movie in motion was disappointing.

so why is this? limit of time/power to implement best algorithms? limit of algorithms developed so far? or is it inherent problem of computer drawings (i.e. it's like asking why a pencil drawing will never look real)

 

[CTho9305]

One of the major factors is called sub-surface scattering. Look here for some info on it. Other factors are "too perfect" lighting and "too perfect" motion.

 

[xts3]

Long story short: Things can only look as good as the math used. So deficiencies in mathematical models are one of the #1 reasons.

There are many reasons. Motion: Modelling physics of motion on a 3D model is very difficult to do and still have it look 'natural' because a 3D models and the details are 'rigid' set of fixed points wrapped in textures. It's hugely a matter of processing power, rendering technology and time and money spent on the art.

CG characters look CG as a function of the state of rendering technology, characters are not composed of discrete atomic units in 3D modelling software, they are essentially fixed point flat 3D objects (triangles/quads/etc) wrapped in textures. So you can't get the subtlety of human motion yet without finding a good algorithms to model it correctly, also due to the rigid body nature of 3D characters there is complex physics that happens when you move your arm that just doesn't effect your arm per se, but also your skin, the muscle beneath your skin, the pumping of your blood through your veins, etc, that all effect the motion of many parts of your bodies skin and sometimes other muscles. We're years away from rendering the types of complex realism you'd see in a real world "iron man" competitions where you see these ripped people with bulging muscles and veins.

The resolution problem: The smallest "detail units" and maximum resolution in a 3D package can't compete with the real world atoms (i.e. think distance detail and clipping problems). If you look at skin of the Final fantasy characters you notice the skin textures and the models aren't detailed enough. If you really pay attention when they are in close the skin on both aki and the guys are totally flat and uniform!, real peoples skin is not flat, its bumpy and pourous and dirty and wrinkly/stretchy, try looking at your skin in the light from different angles having the light reflect off it then push and pull and stretch your skin, then go back and look at CG characters and pay attention if you see any "realistic" stretching. Real peoples skin and arm hair don't look "flat" or "matte" either, if you not notice all the characters in final fantasy have 'the same' uniform look. Real peoples skin asbsorbs/reflects light differently. i.e. hold up a flashlight to your fingers and you see that light passes through your flesh so flesh is transparent and lets a certain degree of light pass through it.

The lighting models and color alogirthms they use, make the skin look not real. They can't seem to compensate for proper 'material' light absorption/etc. As you already know modeling clothes and folding skin is difficult, part of the reason the characters look so plastic and don't animate "right" is because their 'clothes' and 'skin' are super rigid and 'hard body', but this largely depends on how the characters in each scene were modelled, what artists worked on them, and a host of other factors. Some characters clothes are part of the model itself and not an actual seperate object (those "cheating", time and money saving, artists/people don't help!).

It's not an inherent problem of computer drawings per se, its the amount of details big and small your rendering technology can reproduce when compared to the real world. Things in the rendering world have to be mathematically modelled, so things can only look as good as your math. The resolution of the pictures of final fantasy are very high but your eyes resolution and that of the real world is much much higher, approaching infinite detail. So even if CG characters even get to start looking real, they will still look "off" simply due to the fact that the world approaches levels of detail that cann't be accurately captured or modelled yet by current rendering tech.

 

[zetter]

Some CG Images are getting closer to looking real, take a look at the Miss Digital World Contest for some examples, Kaya is one good example.

 

[Shalmanese]

Humans are very good at recognising how other humans look and move. It's a hardwired section of our brains. This means that its far harder to make a convincing human animation than any other type since even slight differences can be picked up.

 

[Sabbathian]

There are many problems standing in the way of realistic renderings.....
Example: if you want to render a pencil on the table.... first you have to
model it, that is easy. Now, rendering..... there are many good renderers
out there, but the one who uses them must know everything about them
to get best results. You must make every single material in the scene as
perfect as it is for real...... then, textures should be also perfect..... just
take a pencil in your hand and look how many details there are.......
OK, you`ve done that... you have to get exact material for the table also.
And the most important thing...... lightning..... mostly used techique to
get real lightning in your scene HDRI (High DYnamic Range Image) lightning.
It sorounds your scene with panoramic picture that emmits light and
simulates real lightning..... but that is not REAL lightning...... not even close.

There are simply too many things that should be considered when making
realistic scene, and it would take way too much to render........

 

[f95toli]

Some movies come close "Final Fantasy" (when the "actors" are wearing armor) and also the first movie on the "Animatrix" DVD. It will improve with time.

 

[BenSkywalker]

in animated movies or clips, you would never be fooled into thinking that the models were real people. they always look like plastic or dolls.

A lot of that may be stylistic, you need to compare the best in attempts at realism when trying to compare CGI to real life. IMO ILM still has the crown here, there are numerous shots in EpisodeII as an example that the overwhelming majority of people totally missed were nigh/entirely CGI, including humans.

One of the major factors is called sub-surface scattering. Look here for some info on it.

Already taken care of in the top tier of CGI, and the 'good' Stanford example still looks pure plastic(very poor).

Other factors are "too perfect" lighting and "too perfect" motion.

The first isn't actually about the lighting itself per se, but they need an added effect there that is still missing(at least, if they are using a proper radiosity solution). The motion isn't too perfect, it is too robotic from the lower tier CGI houses.

CG characters look CG as a function of the state of rendering technology, characters are not composed of discrete atomic units in 3D modelling software, they are essentially fixed point flat 3D objects (triangles/quads/etc) wrapped in textures.

That sounds almost Tron level retro Fixed point flat 3D objects wrapped in textures? Even the houses that don't try for photorealism are far past that and have been for years(Pixar for example).

The resolution problem: The smallest "detail units" and maximum resolution in a 3D package can't compete with the real world atoms

Nonsense, we aren't talking about anything resembling a hologram here, we are talking about the extremely low resolution limits of displays. There are certain factors they need to compensate for, but my digital camera has no problems at all showing me images that don't suffer from a CGI look despite its very low resolution(compared to life anyway).

If you look at skin of the Final fantasy characters you notice the skin textures and the models aren't detailed enough.

Multiple generations ago there were numerous issues that they have since rectified, that is for certain. FFTSW was pretty good, but it had tens of thousands of flat out bad/lazy mistakes that even with their relatively limited technology they could have done much better with.

Comparing the top tier current off line CGI I see their biggest drawback being their complete inability to render air properly. All current render engines treat air as empty non distorting space, every CGI shot gives the appearance that it was 'filmed' at -50 degrees or so on a perfectly clear day with absolutely no wind, even in tropical downpours. There is currently no compensation for humidity and air borne particulates that cause light refractions throughout the air itself. This works on both ends, disperssion of light in the general sense and also the distrortive effect it has on what you see. This more then anything is why observant people can still easily spot CGI no matter how well it is done.

To do a proper calculation on that level it would take years for a Cray per frame, but they should be able to do a decent approximation by figuring out a layered depth filter- a sort of alpha mask with a slight distortive effect for a given distance based on the environmental conditions. This wouldn't be too intensive at all, but figuring out what filter to use per shot and have that calculated out based on Z depth in relation to the camera, and the environmental conditions will be pretty tough.

 

[buleyb]

Originally posted by: BenSkywalker
To do a proper calculation on that level it would take years for a Cray per frame,

Not to go off topic, but the problem isn't complexity on this kind of computer, but poor application of the problem. There are better solutions to this type of problem, and better general purpose CPUs matched with application specific FPGAs in good parallel systems are a better way to go than massive multicomputers all struggling at the same bottlenecks in code...

 

[xts3]

You said: That sounds almost Tron level retro Fixed point flat 3D objects wrapped in textures? Even the houses that don't try for photorealism are far past that and have been for years(Pixar for example).

I think you read that a little too literally. I meant that that models have a problem of being 'rigid' and 'flat' and unmoving (i.e. fixed point, when a characer moves their torso in a game the model doesn't "stretch" at all or in the movies even very well, like a real torso would, its more of a "torso piece" like a suit of armor. Most games today and some movies like final fantasy have the 'solid' object problem and they yes they are composed of triangles/normals with mapped textures for their character models, unless you have any real game examples from the examples I've been using where this isn't the case? I said it is dependent on the rendering technology.

I figure that most of the time we don't want to see "real" special effects we want to see new things that look like they could be real and don't look out of place in a scene.

 

[Matthias99]

Originally posted by: xts3

You said: That sounds almost Tron level retro Fixed point flat 3D objects wrapped in textures? Even the houses that don't try for photorealism are far past that and have been for years(Pixar for example).

I think you read that a little too literally. I meant that that models have a problem of being 'rigid' and 'flat' and unmoving (i.e. fixed point, when a characer moves their torso in a game the model doesn't "stretch" at all or in the movies even very well, like a real torso would, its more of a "torso piece" like a suit of armor. Most games today and some movies like final fantasy have the 'solid' object problem and they yes they are composed of triangles/normals with mapped textures for their character models, unless you have any real game examples from the examples I've been using where this isn't the case? I said it is dependent on the rendering technology.

I figure that most of the time we don't want to see "real" special effects we want to see new things that look like they could be real and don't look out of place in a scene.

There's a BIG difference between the simplistic methods used in today's 3D gaming hardware and the raytracing/radiosity rendering systems used for things like feature film CG animation. You seem to keep flipping back and forth between them without distinction, which is probably why people are getting confused here. The limitations you're discussing in that paragraph don't really apply to more recent professional animation projects; high-end skeletal animation packages DO take into account things like muscle and skin movement/deformation.

You really have to draw a line between the hardware-accelerated (but still inherently polygon-based) modelling that you get on a PC's 3D hardware and the much more complicated things you can do with modern software rendering packages (such as Maya, Alias Wavefront, RenderMan, etc.)

 

[Oyeve]

Some of those CG movies can take months or over a year to render and compile. Realistically it can be done but the time constants are also a major factor.

 

[BenSkywalker]

I think you read that a little too literally. I meant that that models have a problem of being 'rigid' and 'flat' and unmoving (i.e. fixed point, when a characer moves their torso in a game the model doesn't "stretch" at all or in the movies even very well, like a real torso would, its more of a "torso piece" like a suit of armor.

For games or low end/outdated CGI perhaps, but otherwise no.

Most games today and some movies like final fantasy have the 'solid' object problem and they yes they are composed of triangles/normals with mapped textures for their character models, unless you have any real game examples from the examples I've been using where this isn't the case?

Talking about games when we are having a discussion about CGI is like bringing in the performance of a Yugo when we are talking about top fuel cars, very disjointed particularly when you mix them together. They are both cars, but should not be compared in any realistic way.

For what we are talking about spline based rendering is far more common then fixed point geometry(although when rendering an inorganic non deforming model it makes more sense to use basic static vertices as they are less system intensive).

I said it is dependent on the rendering technology.

If someone asks can you play a DVD on a PC using software decoding I don't assume they are running an 8086.

I figure that most of the time we don't want to see "real" special effects we want to see new things that look like they could be real and don't look out of place in a scene.

If CGI based special effects are done good enough, you won't know you saw them at all. That is what this thread is looking at, what that will take.

 

[Oyeve]

If CGI based special effects are done good enough, you won't know you saw them at all. That is what this thread is looking at, what that will take.[/quote]

Plain and simple answer. It can be done. You need a super CPU(s) power and a lot of time.

 

[hahher]

Plain and simple answer. It can be done. You need a super CPU(s) power and a lot of time.

since everyone expresses sentiment along these lines, does there exist just a short video clip (~1min) showing off what's possible?

 

[Oyeve]

Originally posted by: hahher

Plain and simple answer. It can be done. You need a super CPU(s) power and a lot of time.

since everyone expresses sentiment along these lines, does there exist just a short video clip (~1min) showing off what's possible?

Yes, although I cant remember right now. I have seen some and they choked even on my system (check my rigs). I will see if I can find them and get you a link. Amazing stuff.

 

[EarthwormJim]

I think one of the problems with CGI is not the technology behind it, but the situations it is implemented in. It seems almost as if CGI is thrown in to shock and awe the spectators, not to simulate reality. Rendered characters often do things that normal people cannot do, camera angles do the same too. The effect is often a rubbery or machine like character, like for example Spiderman from the movie. Impossible camera movements and angles just scream CGI to the audience, but if done with traditional camera movement and angles, the audiance may not even have realize it's all computer generated.

 

[0roo0roo]

not to mention final fantasy was a very poor example. at that time they didn't have the power to fully model the underlying muscles of the humans, so really they were just shells, and looked it. the facial animation was simply hand done, it was very primitive. although they had motion capture, the bodies were just action figures, they did move in the subtle ways a human does. its just a matter of technology and time. we are still quite primitive. stuff today is rendered as real as we can afford it, not as real as it can be.

 

[Shalmanese]

www.archive.org has the siggraph winners from a few years back I think. you can download most of them for free. Apart from that, "The flight of osirus" was a short clip in the animatrix which was rendered by square and has some fairly advanced stuff. Final Fantasy still hasn't really been surpassed yet AFAIK simply because nobody can put that much money into rendering stuff.

Of course, this is all purely CG. If you want assisted CG, crowds are routinely generated from a few extras, the entire crowds in the titanic were CG based. most likely the reason you never notice it is because its so seamless.

 

[jhu]

this look real enough?

 

[BigPoppa]

Originally posted by: jhu
this look real enough?

No. Its pretty...blah. Whats so special about it?

 

[Oyeve]

Originally posted by: jhu
this look real enough?

Buildings are ok, the cars are definately not. Also, is this rendering movement or a still?

 

[Matthias99]

The person in the lower right has been motion-blurred, it seems, but it's hard to tell the true quality, since the images there are obviously compressed JPEGs (you can clearly see the aliasing artifacts around the edges in the 'super' version).

Looks pretty damn good to me, though. Volumetric (probably radiosity-based) atmospheric lighting, depth of field effect on the background, good handling of the water on the pavement... pretty good composition, too. At first glance you might think it was a photograph.

 

[jhu]

here's a larger image. on a quick glance, it looks like a real picture.

 

[Matthias99]

Still a compressed JPEG; this one looks like it had the color depth reduced (look closely at the sidewalk in the foreground; you can see speckling that's not in the ones you linked to before). What you really need is a .PNG or .TGA at about 1200x1600 (or higher); that would show off the detail work. However, you probably won't find one unless you have the print and can scan it in yourself; few people would buy it if you could just download a full-quality version of it from somewhere.