Hacker News Reader

Comment by wokwokwok | Hacker News Reader

Did you see them putting the black panels up in the video and measuring the color spectrum off the costume, and covering the skylight?

They did that to get a 'perfect' mask.

If you use infrared, which is generated by random environmental objects, how would you prevent random spill?

Practically (my understanding) is if you got random spill on the surface, you'd get random alpha in the mask.

The point of using the sodium light is that it's a single wavelength that you really really would not get naturally, so you can totally control the emission sources of it.

Could you do that with a single frequency infrared? I guess it's possible, but I'm not sure how you'd do it technically, and it's unclear what the benefit of using a different frequency would be?

View on HN

Near infrared isn't generated by random environmental objects. You're thinking of far infrared. Infrared is a big spectrum, much larger than the visible spectrum.

It may be generated by lights but could be filtered out. It is generated by the Sun, but so is the sodium wavelength. The technique would work indoors just as well as the sodium one, but with important advantages. I think it could even work pretty well outdoors if you used a wavelength that is mostly filtered out of sunlight by the atmosphere, or maybe two different wavelengths in two channels simultaneously.

Huh, my bad, I thought thermal sources (ie. like random hot stuff, not glowing hot stuff) emitted a wide band of frequencies including near-IR, but fair enough. You'd still get it from any sort of incandescent light source right? (same as a sodium lamp)

> but with important advantages

What's the benefit over using any other single frequency visible light?

Even if the atmosphere absorbs most of a particular band, if any of it gets through, your mask is going to be messy, which kind of ruins the point; I don't know anything can really fix that for outdoor purposes?

I mean, broadly speaking, all you need is a way to emit a sharp frequency band and avoid having any environmental pollution in the same band.

If there's a good way to do that in near-IR, it would work; but I feel like the same could be said about pretty much any frequency.

I'm not aware of anything specific about near-IR that makes it particularly attractive.

(but, clearly, I'm no expert in near-IR)

The advantages I stated were

> It would allow using full spectrum lighting, and would be easier to selectively filter.

Making a broad spectrum IR filter is far easier than a notch filter for a single visible wavelength. In general it's harder to make more selective filters and an IR filter can be much less selective because the IR band is so large and far away from any visible light.

Also the infrared version would allow using any color background including black, and the on set lighting would appear to the eye exactly as it does to the camera. With the sodium vapor process the bright yellow background would make everything on set look different in person because the eye wouldn't be able to subtract it the way the camera does.

Yes, you would need a filter on any incandescent light. But this is also true of the sodium vapor process. Actually I'm not sure anyone makes a sodium vapor wavelength notch filter suitable for putting in front of an incandescent light. Consequently, as seen in the video, they restricted themselves to special LED lights that emit essentially only three wavelengths. These are very much not full spectrum and will not produce natural colors for some materials. And they had to completely block all windows, whereas for IR you could just install some IR blocking film on the windows and still use natural sunlight.

The whole point of this technique is that you have that sharp frequency band and a crystal / filter that splits like band out.

You're basically proposing something completely different here.

So... while I agree you could do something as you describe, and I'd be interested to see the results of it, I'm not sure what the results would be like.

> The advantages I stated were

>> It would allow using full spectrum lighting, and would be easier to selectively filter.

All I can say is the CC results in that video were pretty great, despite the obvious caveats, compared to green-screen.

If you (or anyone) can build on it to do better, people will be interested in it; but there's obviously a difference between us idly speculating on how/if it might work, and this, where they've actually done it and shown that it does work.

> "they do, it's just uncommon and you haven't heard about it"

I've never heard of it being done.

If anyone has, please post a link or something.

No, the point of the technique is you have a frequency band that you can isolate and use in the background and not in the foreground. Using IR instead of visible as your isolated band makes the isolation easier and cheaper with fewer artifacts and limitations. It's still a very similar process otherwise.

Yes, Corridor's results were great. Fewer limitations would make the process even better!

Near ir LEDs are also cheap and easy to power, with its what's in remote controls. Low pressure sodium lights aren't. I remember using them in postgrad labs and it wasn't fun.

Alternatively you could use a solid state laser or narrow band LEDs if you need the band to isolate to be in the visible spectrum too.

> What's the benefit over using any other single frequency visible light?

One advantage might be to pulse the IR and detect it via that, if the detectors are sufficiently fast. Maybe have a filter that's pulsed also (Pockels cell?)

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

One problem you might run into is that a lot of common plastics are opaque to NIR light, so you might find certain materials gave you strange results (water bottles that appear transparent to the eye would not actually pass the NIR light needed to make the mask layer).

Reply on news.ycombinator.com