HD For Indies

So the specs and pricing and stuff are up for the RED camera, and let's just say off the bat that it rocks. But there are SO many choices in shooting modes it can be confusing. There's a matrix on the handouts here at the show, and perhaps on the Red site.

So here's my take on it:

More on RED

The camera was designed to modular - start light and small so it'll go places big cameras won't, but be able to bulk it up for large production needs using the optional cages that give hard mount points for rods, lens support, viewfinders, remotes, etc. It is a helluva lot easier to make a small camera big than a big camera small.

It has more in common with the working style of a film camera than it does a traditional video camera - the ability to shoot anywhere from 1-120 frames per second, between 720 and 4520 pixels tall.

A bit more on recording modes: there are four variables here:

1.) What pixel size you shoot

2.) What pixel size you record

3.) What video/data format are you recording

4.) what physical recording device does it go on.

One at a time:

1.) What pixel size you shoot -

1.) The Mysterium Sensor is 4520x2540 pixels, and is a Super35mm sized single CMOS sensor. All modes start from there.

2.) So when shooting in the following modes, you're recording:

4520p - it is the full output of the sensor, up to 60 fps

4K - it crops off around the edges to get down to 4096x2160, up to 60 fps

2K - in this mode, up to 60 fps (more on other modes in a minute) - start with the 4K above, and then scale it down - so it is super sampled (or oversampled, if you prefer that lingo) to give smoother results with less noise, better signal/noise ratio, less aliasing, and this in turn helps the dynamic range. So when recording 2K or less, you have an oversampled, cleaner/smoother image. Remember this applies to the rest of the formats below

1080p - take the 2K and crop it a bit more - so for 1080p, you have an image that is oversampled about 4x - it is 2x in both the horizontal and vertical dimensions. This is GOOD.

1080i - take the 1080p and use each 1080p60 frame to generate a 1080i field. This is, yet again, oversampling, and provides a VERY nice, clean, sharp, smooth 1080i. This is actually an ideal way to generate a 1080i image and is in no way a compromised image

720p - through a combination of scaling and cropping, the 4520p is scaled and cropped to 720p for a highly oversampled (I think about 3x in each dimension, I'll need to ask Graeme Nattress to confirm), so should be extremely clean. This description differs slightly than the handout on the show floor, but Graeme assures me 720p is optimally derived after I grilled him on it, not just scaled down from the 1080p as may be indicated elsewhere.

In the viewfinder, whatever mode you're shooting will show with reticles around the SurroundViewTM viewfinder - you'll be able to see what's in frame and what's about to come in frame - more like an optical viewfinder on a film camera.

OK, back to 2K - what if you want to shoot some serious high speed, like 120 fps? And what's this about the ability to use Super 16mm lenses?

It works this way -

The sensor is Super35mm sized. The camera can accept Super 16mm lenses. when you mount a S16 lens on it, the image hits the sensor the size of S16 film, not the size of S35 film from an S35 lens. You're now using a small center piece of the imager - this is called windowing. By windowing the sensor, you can use S16 lenses and get a 2K res image out of the camera, you're just not oversampling anymore. But, by using a smaller piece of the sensor, it is possible to crank up the cycle rate on it, and get a higher frame rate. Therefore you can go up to 120 fps at 2K res (or 1080p or 720p, and 720p WOULD be oversampled in this case) when using S16 lenses. And we're talking about DCI 2K here - 2048x1080, not the anamorphic film scan 2048x1556 pixels.

The decision was made to keep the depth of field the same for 1080p and 720p, FYI, so it doesn't change when you switch modes. Convenient! You're NOT windowing when switching between those modes.

A word on anamorphic - you don't need to use anamorphic lenses with this camera (unless you really want to). I had a lengthy chat with both Graeme Natress and Stuart English about this, and they said what we have here will work just fine for widescreen implementations. I'll probably follow up in depth and write more about it, but in general, you're in good shape - it is a native 16:9 sensor to start with, and you can always crop the image to get 2.35:1.

3.) What video/data format are you recording in? It depends:

When shooting 2540p - you're assumed to be shooting RAW at this point and want maximum quality. RAW is the unprocessed direct output of the Bayer pattern from the CMOS sensor. It is uncolor corrected, and isn't even directly viewable - it's still grayscale data at this point that has to be processed to figure out which pixels are R/G/B. In any case, 4520p is always RAW.

When shooting 4K, same thing.

When shooting 2K, you can now start using the REDcode codec. For 2K, it is assumed you want log data (like a film scan or Viper FilmStream mode), which does a better job of recording highlights and is akin to 12 bit linear. If that doesn't make any sense, that's OK, just know that in 2K you have a choice - either RAW, or REDcode 10 bit RGB 4:4:4 log variable bitrate wavelet. It might be possible to go out the dual link HD-SDI with 2K, I'll have to check. In theory it is possible, especially with the new AJA Kona 3 card's capabilities.

When shooting 1080p, you have yet more choices - you can record uncompressed out the HD-SDI (single or dual link), or you can use the REDcode 10 bit 4:2:2 codec. In 1080p, I THINK you can do RGB 4:4:4 1080p, but I'll need to double check - that isn't explicitly stated on the data sheet. But 10 bit 4:2:2 REDcode wavelet is an option, as well as uncompressed to the REDRAID.

There seems to be a gap here that I need to follow up on - is there a 10 bit log RGB 4:4:4 for 1080p? If there a 10 bit linear RGB 4:4:4 for 1080p? Or is it assumed you'll start with a 2K and crop down in post somewhere/somehow? I'll have to ask about that.) - actually, I just talked to Graeme about it - yeah, it'll be fine. He said something about codecs are still under development, and Red won't do anything to artificially limit users. My own thought - yeah, that would be so completely against the spirit of what they are trying to do here.

For 720p, 10 bit 4:2:2 to REDcode VBR wavelet codec, out the HD-SDI single link uncompressed, and possibly uncompressed to REDRAID, again I'll need to check the matrix on that one.

And lastly, where do you record it to?

Options include:
RED-RAID - high speed RAID for RAW recording, possibly uncompressed recording as well - have to check. This is an external device, they are considering either Infiniband or dual link fiber channel of some sort.
RED-RAM - I'm guessing for short bursts of RAW or high speed I'm guessing, external as well. But it'll have a higher capacity than REDFLASH. Sufficient.
REDFLASH - internal 32-128GB flash memory. I need more info, but I'm guessing for shorter takes to REDcode, maybe RAW, maybe uncompressed?
RED-DRIVE - internal to camera, hard drive based (probably 2.5" drive mechanism)
...or out the HD-SDI, single or dual link depending on frame size and rate and what fits in the HD-SDI spec for frame and data rates and bit depth. At that point, you can hook up to conventional decks or DDRs or other recording gear.

If you assume a bitrate of 100 mbits, and assume a REDDRIVE of at least 40 GB (and these are just safe assumptions about datarate and drive size for an example here), that'd be an hour or so of footage. Higher data rate and higher capacities obviously change that. But count on being able to probably do about an hour of recording on the REDDRIVE.

Again, that is all just to say this camera is all about choices and finding what works best for you.

-mike

Labels: Red