Some Eye-Popping Research From Siggraph

jamie found links to a discriminating selection of Siggraph papers at waxy.org. Among the more captivating: automatically improving the attractiveness of faces in portraits; automatic substitution of similar faces into photographs (with potential applications such as a privacy-enhanced Google Street View); and using still photographs to enhance video of a static scene.

Comments

FX Show Nip/Tuck said it best

[Anonymous Coward]

"Beauty is Symmetry, and you have none"

One of the main characters in the plastic surgery show Nip/Tuck made that comment. It seems as if TFA applies said comment.

Re:FX Show Nip/Tuck said it best

[debuglife]

This is a typical slashdot comment. Uninformed, belittles others work. Symmetry might work in some pictures. It might not work in others. Besides, the feature extraction needed to establish symmetry is non trivial (its the same as what the author of the paper does). If this was so trivial, why didn't you build a system like this. The world belongs to the doers, not the "Ah, I told you so" crowd.

So in summary

[mrbah]

Just add symmetry and make thinner.

Re:So in summary

[jacquesm]

that was exactly what I was going to write :) But the first guy sort of defeats that rule, he's actually a little broader in the face than the source image.

A simpler rule would be 'add symmetry', mirror the left half of the face (or the right half, flip a coin).

Adding a smile also goes a long way towards making people prettier, in fact a smile really is the best make-up.

Re:So in summary

[Anonymous Coward]

The algorithm seems to be highly dependant on gender. For example women aren't allowed to have a broad chin, men aren't allowed to have a narrow chin. So what it does is enforce gender stereotypes, which is probably what the majority base their perception of beauty upon. Sad but true.

Re:So in summary

[Jorophose]

Not add symmetry.

Follow the golden rule.

Ghost of Clarke seen skulking nearby.

[Purity Of Essence]

The first two are meh-worthy, but the last one approaches magic-grade technology. Wow!

Re:Ghost of Clarke seen skulking nearby.

[mrbah]

Microsoft spends billions of dollars researching things like that, but never brings any of them to market. Look at the "Image Deblurring with Blurred/Noisy Image Pairs" paper -- it's a marketable, easy to use technology that would be of huge benefit to typical consumers, yet chances are good it will never be commercialized. Contrary to popular opinion Microsoft does innovate, it's just that all the good stuff gets killed by some committee full of assistant senior project project team manager manager mangers.

Re:Ghost of Clarke seen skulking nearby.

[jagdish]

Now if they integrated all this tech with a camcorder, everyone could have a device with video recording capabilities several times that of current HD recorders.

Re:Ghost of Clarke seen skulking nearby.

[zalas]

-- it's a marketable, easy to use technology that would be of huge benefit to typical consumers, yet chances are good it will never be commercialized.

I've noticed that a lot of SIGGRAPH papers will either only work for a small subset of inputs you would want to feed it or need a properly controlled environment to work and might need a lot of tweaking to get looking correctly. In my opinion, SIGGRAPH papers depend on demonstrating excellent best cases while mainstream consumer products require that the worst case be also acceptable.

real footage?

[apodyopsis]

the question is in twenty years time will you trust the news you see on TV?

when cheap, easy, video editing allows this then supposedly real footage: news, family videos, wedding snaps will lose all veracity.

after every girl wants to look good for her wedding...

and before somebody says "it will never happen" this is only a logical extension of red-eye removal.

Re:real footage?

[pembo13]

I get the feeling from you that you trust it now. I find this confusing myself. Considering that an apparently large portion of Slashdotters very much consider themselves rationalist who do not believe things without proper evidence, it seems weird to me that many simply believe what they see in the news. These past week (maybe 2) there were at least two cases circulating around the internet where it had been observed that CNN has used footage from one event, trying to pass it off as that of another event. And that's pretty low tech.

News reports should be only be as trusted as logic can be applied to the report.

Take for instance the recent story of a Russian sipper shooting at a reporter. A few questions came to my mind:

• What kind of sniper takes such a shot and misses?
• What kind of sniper misses, and doesn't take a second shot?
• How does one tell the affiliation of a sniper? Do they sign their bullets or something?

News stories should be treated as untested pieces of evidence -- in most cases at least. The advancement of technology will only make it more difficult to tell truth from fiction.

Re:real footage?

[owlnation]

the question is in twenty years time will you trust the news you see on TV?

You trust it now? Are you new here?

Re:real footage?

[rossz]

> What kind of sniper takes such a shot and misses?

No one is perfect. Long distance and wind variations can cause a miss.

> What kind of sniper misses, and doesn't take a second shot?

The smart sniper. There was no way a second shot would have hit. Everyone was moving around too much.

> How does one tell the affiliation of a sniper?

If they shoot at you, you can be sure it's the enemy. The sniper would have easily figured out which side the potential target was on.

> Do they sign their bullets or something?

Signing the bullet would have screwed up the ballistics. Snipers are extremely anal retentive when it comes to their rounds. They usually use hand loads and they buff the round to remove any imperfections.

FYI, a close friend was a sniper for SpecOps.

It rather looks like an uglyfication.

[Hurricane78]

I could not decide, which versions looked better. I only recognized that they ere mostly non that good looking on both fotos.

Then I looked closer, because I know a bit about the methods behind it. And they did some big errors, like copying the one side of a face to the other, when the face did not look perfectly straight into the camera. This gave some weird results. Some faces even looked quite unnatural (especially, but not solely the focus on huge foreheads.

P.S.: I'm happy that I now since the last months know trough experience, that my opinion that looks matter, were wrong.

does anyone realize the implications of this tech?

[DragonTHC]

Video does lie now.

Can video ever be trusted again where evidence is concerned?