Pricing F5 F55

[starcentral]

Anybody who thinks of RAW that way will never ever benefit from its greatest strengths. It's hardly a point and shoot excuse. Just like anything else...there's a right way to use it to benefit from it the most.

You (and several others here) have completely taken what he said out of context. In many ways RAW is point and shoot, everyone knows shooting digital and lighting for it is harder than for film, shooting RAW however brings some of that back some of the lattitude you had with film.

But shooting to very wide-dynamic range Log gives you almost the same safety buffer as RAW.

This is incorrect. The higher the dynamic range of the camera the less lattitude you are going to get with lower bit-depth based codecs when it comes to grading ie. 10-bit. Remember LOG is a curve with giant knee near the top compressing your highlights. You will have lots of room in post for mid-tones but not highlights. With RAW linear it's equal throughout the entire dyamic range of the camera. This is the power of RAW, because you can compress/expand your own highlights in post with more lattitude. With LOG its already compressed and you don't have the same amount of bits when expanding it as with RAW.

Also this discussion is VERY relevant to PRICING because I've read many people justify buying the F55 who say they don't need RAW and will just shoot internally in 4k at 10-bit - but what I'm questioning how great that will be on a 14 stop dynamic range camera, and I'm also saying to really maximize the benefit in post I'd rather shoot RAW on the F5 in 16-bit for less money.

 

[starcentral]

The Avengers was graded from 10-bit Log DPX files - I know because I worked on a bunch of VFX shots for it for the trailer.
The Dictator was graded from 10-bit Log DPX files - I delivered VFX for the feature (and designed his beard!)
Prometheus was graded from 10-bit Log DPX files - I asked the guy who graded it.
Underworld Awakening was graded from 10-bit Log DPX files - the guy who graded it said so.

But shot in RAW:

 

[bruceallen]

^ This is incorrect. The higher the dynamic range of the camera the less lattitude you are going to get with lower bit-depth based codecs ie. 10-bit. Remember LOG is a curve with giant knee near the top compressing your highlights. You will have lots of room in post for mid-tones but not highlights. With RAW linear it's equal throughout the entire dyamic range of the camera. This is the power of RAW, because you can compress/expand your own highlights in post with more lattitude. With LOG its already compressed and you don't have the same amount of bits when expanding it as with RAW.

If you had a theoretically perfect camera, sure.

But the actual signal-to-noise ratio of all digital and film-based cameras is not that great that it becomes a problem.

It definitely has worked well as a standard for film scans for coming on 20 years - and film has plenty latitude.

The Cineon spec puts pure white at value 685. Values above 685 were reserved for specular highlights. You have like another 340 values left to describe just the specular values... that's a lot.

If digital had greater latitude than film, then we'd need a new standard. But it doesn't yet.

Of course, for HDR imagery this is not enough... but I'm pretty sure the F55 isn't an HDR sensor.

This discussion is VERY relevant to PRICING because I've read many people justify buying the F55 who say they don't need RAW and will just shoot internally in 4k at 10-bit - but what I'm saying is that won't be great on a 14 stop dynamic range camera and to really maximize your benefit in post I'd rather shoot RAW on the F5 in 16-bit for less money.

Agree! Ultimately I think we just need to do tests... but seriously, take some Red footage, set the output to RedLogFilm in REDCINE and try it out. If you set your ISO / contrast settings quickly to a where black is somewhere around 100 and white non-specular objects are in the 600 range... I really don't think you lose much at all in terms of useful information versus the RAW. That's just my personal experience.

Bruce Allen
www.boacinema.com

 

[starcentral]

Avengers was shot in RAW to ARRIRAW (12-bit bayer) according to Seamus McGarvey...

 

[bruceallen]

Avengers was shot in RAW to ARRIRAW (12-bit bayer) according to Seamus McGarvey...

That is true. Then they made 10-bit log DPX files from that (before grading) and distributed it to VFX.

Seriously, what is the difference between:

1. Shoot 12 bit RAW in camera -> convert in post using generic settings (basic WB, exposure) to 10bit Log -> grade 10bit Log
and
2. Convert to 10bit Log in camera using generic settings (basic WB, exposure) -> grade 10bit Log

?

In terms of latitude they should be largely identical unless the in-camera processing making the 10bit Log file totally sucks or you need to make insane changes to WB and exposure in post.

Bruce Allen
www.boacinema.com

 

[Richard Allen Crook]

This is incorrect.

Not so fast. Ive had this discussion so many times. Raw (not RAW...it's not an acronym) is a property of the digital signal after the sensor has recorded light. Dynamic range is a property of the sensor. The more information the sensor acquires (dynamic range) holds more weight as to what we can alter in the grade than whether that info is raw or log gamma. Raw doesn't mean it can gain back more information than what the sensor can't see.

 

[Grug]

That is true. Then they made 10-bit log DPX files from that (before grading) and distributed it to VFX.

Seriously, what is the difference between:

1. Shoot 12 bit RAW in camera -> convert in post using generic settings (basic WB, exposure) to 10bit Log -> grade 10bit Log
and
2. Convert to 10bit Log in camera using generic settings (basic WB, exposure) -> grade 10bit Log

?

In terms of latitude they should be largely identical unless the in-camera processing making the 10bit Log file totally sucks or you need to make insane changes to WB and exposure in post.

Bruce Allen
www.boacinema.com

If they're not actually doing any processing of the raw files themselves (i.e. converting them directly to DPX), what reason would they have shot raw for in the first place? Wouldn't recording directly to DPX on something like the Gemini be of equally high quality and save the additional time in post?

 

[bruceallen]

If they're not actually doing any processing of the raw files themselves (i.e. converting them directly to DPX), what reason would they have shot raw for in the first place? Wouldn't recording directly to DPX on something like the Gemini be of equally high quality and save the additional time in post?

Well, not "no processing". My guess is basic processing - eg white balance and basic exposure / basic LUT (for camera, not for every scene).

My speculation:
1. Not sure if anything existed that could decode ARRIRAW to 2880x1620 DPX files in realtime in the first place.
2. Shooting ARRIRAW saves a lot of disk space versus shooting 2880 x 1620 DPX (which stores R,G,B per pixel, versus just Bayer data) and can decode to the same quality later.
3. Unlike REDCODE, ARRIRAW is not much hassle to decode later, as there is no wavelet decompression stage and only a quick de-Bayer.
4. Maybe nice marketing for Arri?
5. Maybe the post companies already had a nice workflow sorted out for ARRIRAW files?
6. Comfort of knowing you are at MAXIMUM CAMERAS SETTINGS YEAAHHH!
7. Makes RED fans shut up.
8. Possible actual slight quality difference. You would run a 3D LUT on the file to normalize it to 10bit Log finishing space and that LUT would work a tiny bit better if given a 12-bit linear source?

In either event, all that was needed after VFX (and before the grade) was 10bit log DPX files, at slightly-above-2K resolution (to account for 3D conversion).

The whole point is that 10bit Log DPX 2K is just the standard recipe for what you put into a film finishing pipeline.

10bit Log DPX has handled the latitude and color range of film with few problems for nearly 20 years.

Please ask yourself whether you really think that digital exceeds the latitude and color range of film before proclaiming 10bit Log to be inadequate.

Bruce Allen
www.boacinema.com

 

[dracalat]

Bruce, thank you for your excellent posts and 'demythologising' Raw. I often try to explain all this to clients, but some insist and swear vehemently on shooting Raw. My eyes simply don't see any difference between Alexa Log c and Arri Raw (but hey, maybe I'm just getting old).

 

[Kolor-Pikker]

Haha, looks like my point & shoot comment stirred the pot up somewhat, I guess I was a little too flippant. Obviously it's not it's not that simple, and I've used RAW for a very, very long time on many cameras, so I know what it's for.

I don't think anyone mentioned this yet (TL;DR), but Raw actually requires quite a bit of processing power to actually generate an image (has to be calculated from scratch), which is why uncompressed RGB formats like DPX are used. It's not about disk space, disk space is ridiculously cheap in the grand scheme of things, it's about processing, and no one is actually going to grade the raw Raws. It takes up to 20 seconds for me to convert a still 5k equiv. frame from a stills camera using a high-quality Raw converter such as Capture One, do you really think it's feasible to process entire clips of footage and grade them at the same time at maximum quality? Even with the souped-up hardware that you might expect in post houses, it's a very serious load on the hardware, a 20TB array is better than extra 12-core computers just to churn through the data. On the other hand, with a proper disk array, 10-bit DPX is very fast.

Now, a fact: 10-bit is enough for anything within practical limitations.

Let me clarify: The stock would have to be utterly clean of noise in order to have a real 14-stops of DR, like computer graphics clean. The noise floor of a sensor is always fixed relative to each stop, but sensors are always linear devices by nature, this means each stop of light hitting the sensor registers half as much levels as the one from the top, therefore, assuming linear 12-bit encoding:

-Highlights-
Stop 1: 2,048
Stop 2: 1,024
Stop 3: 512
Stop 4: 256
Stop 5: 128
Stop 6: 64
Stop 7: 32
Stop 8: 16
Stop 9: 8
Stop 10: 4
Stop 11: 2
Stop 12: 1
-Shadows-

A fixed level of noise could be relatively low compared to the highlights, but very high compared to the lower tones. Of course, it's not possible to have 14 stops of DR in a linear 12-bit space, because bit-depth is directly proportional to the amount of stops a sensor can record, so that's where Log comes in. The Alexa uses dual-gain and 16-bit ADC to increase the dynamic range of the camera and get a clean 14 stops by truncating the shadows and then fits it's DR into 12-bits by using Log, which compresses the highlights in... but this is an exception allowed by recent technology, in realty, noise dithers shadow detail to the point where it can safely be represented by as little as 10 bits of data.

Once properly logged to 12 or 10-bit, most tones in the image will have an equal amount of steps assigned to them, but not a moment before that. In all, 10-bit assigns about 70 gradations per stop, more or less.

Hope this clears some things up.

 

[Admit One]

It's not about disk space, disk space is ridiculously cheap in the grand scheme of things, it's about processing...

Different de-Bayer algorithms produce different images. You will even get a different image with a Red-Rocket card than with Redcine-X.

This is why it is best to de-Bayer in post, rather that in the camera where very limited on-set time, and camera processing power is available.

Since we constantly demand more pixels/depth/fps, this will continue to be the case.

Once de-Bayered, the choice of RGB storage (linear/log,10/12/16bit) is somewhat less important, and ideally should fit the project.

Cheers.

 

[starcentral]

10-bit = 300 gradations per stop?

8-bit = 256 steps
10-bit = 1024 steps
12-bit = 4096 steps
16-bit = 65,536 steps

 

[Kolor-Pikker]

Different de-Bayer algorithms produce different images. You will even get a different image with a Red-Rocket card than with Redcine-X.

This is why it is best to de-Bayer in post, rather that in the camera where very limited on-set time, and camera processing power is available.

Since we constantly demand more pixels/depth/fps, this will continue to be the case.

Once de-Bayered, the choice of RGB storage (linear/log,10/12/16bit) is somewhat less important, and ideally should fit the project.

Cheers.

Yep, you debayer in post, and the choice of RGB storage is less important. What I was saying is that Raw footage can't be treated the same as RGB, regardless of how you compare bit depth and so on, especially since bit depth behaves differently on a hardware and software level.

10-bit = 300 gradations per stop?

8-bit = 256 steps
10-bit = 1024 steps
12-bit = 4096 steps
16-bit = 65,536 steps

My bad, I was thinking about 12-bit, it's what happens when you're constantly editing. It's actually 1024 / 14 = 73 gradations per stop for 14 stops of DR. But numbers don't matter, the point is noise and the nature of sensors makes 10-bit perfectly fine for RGB storage.

 

[starcentral]

Raw footage does require intense CPU processing since it is not performed in-camera in real-time using a dedicated hardware processor to produce a YCrBr based image format. However I'm a little suprised to hear about disk space?!

RAW is actually more efficient than YCrBr based codecs isn't it? Which is why the FS700 can transmit 4k 12-bit raw over single 3G SDI port - how many ports do you need when using YCrBr based codecs again?

Another way of looking at it (based on 2k images):
• 16-bit RGB raw file with 3:1 compression on RED is around 200mbps
• 16-bit RGB raw file with 5:1 compression on RED is around 100mbps
• Apple ProRes 4444 is around 380mbps
• Apple ProRes 422 HQ is around 250mbps

4k images:
• 16-bit RGB raw file with 3:1 compression on RED is around 780mbps
• 16-bit RGB raw file with 5:1 compression on RED is around 474mbps
• Sony SR master SQ 10-bit 444 RGB is 440mbps
• Sony SR master HQ 10-bit 444 RGB is 880mbps
• Sony SR-Lite 10-bit YCrBr is 220mbps

So Raw shall save your disk space, not eat it.

10-bit = 300 gradations per stop?

8-bit = 256 steps
10-bit = 1024 steps
12-bit = 4096 steps
16-bit = 65,536 steps

My bad, I was thinking about 12-bit, it's what happens when you're constantly editing.

Right and shooting 12-bit is not an option on the F5/F55 which is why earlier I said people might want to reconsider when they justify buying the F55 without ext. raw recorder because it has 4k XAVC internal codec which is 10-bit. I hear it all the time lately, "I'll just buy the F55 and shoot 4k internal and won't need the AXS recorder". For less money you can shoot 16-bit raw on the F5 and end up with a lot more lattitude in post.

To each his own but I'm trying to point out some misnomers about bit-depth and raw vs log.

 

[Kolor-Pikker]

Raw footage does require intense CPU processing since it is not performed in-camera in real-time using a dedicated hardware processor to produce a YCrBr based image format. However I'm a little suprised to hear about disk space?!

RAW is actually more efficient than YCrBr based codecs isn't it? Which is why the FS700 can transmit 4k 12-bit raw over single 3G SDI port - how many ports do you need when using YCrBr based codecs again?

Another way of looking at it (based on 2k images):
• 16-bit RGB raw file with 3:1 compression on RED is around 200mbps
• 16-bit RGB raw file with 5:1 compression on RED is around 100mbps
• Apple ProRes 4444 is around 380mbps
• Apple ProRes 422 HQ is around 250mbps

4k images:
• 16-bit RGB raw file with 3:1 compression on RED is around 780mbps
• 16-bit RGB raw file with 5:1 compression on RED is around 474mbps
• Sony SR master SQ 10-bit 444 RGB is 440mbps
• Sony SR master HQ 10-bit 444 RGB is 880mbps
• Sony SR-Lite 10-bit YCrBr is 220mbps

So Raw shall save your disk space, not eat it.

You must have mis-understood me, I said that RGB data consumes more space and requires more throughput than Raw, but space and bandwidth are easier to manage than the extra processing power that grading Raw data would require. (Edit: up to point though; capturing, processing, grading and releasing a feature film entirely in 4K as in Girl with Dragon Tattoo is fairly complicated in itself, mostly content is captured in 4k is scaled down to 2K dpx for DI)

Right and shooting 12-bit is not an option on the F5/F55 which is why earlier I said people might want to reconsider when they justify buying the F55 without ext. raw recorder because it has 4k XAVC internal codec which is 10-bit. I hear it all the time lately, "I'll just buy the F55 and shoot 4k internal and won't need the AXS recorder". For less money you can shoot 16-bit raw on the F5 and end up with a lot more lattitude in post.

To each his own but I'm trying to point out some misnomers about bit-depth and raw vs log.

Maybe 10-bit is enough for those people, they want the resolution, but not necessarily the latitude Raw offers. Also, you can edit the 10-bit file right away, but the Raw will need processing. FilmDude summed up what I was trying to say pretty well below.

 

[FilmDude]

well here's the thing... I'm one of the people who said I want the F55 to shoot internal 4K and my reasons are the following

I already own a 4K raw RED system..

My clients most of the time want 4K but then passout when then hear how to handle it all in raw format

but the biggest reason is simple. I want an actual 4K camera to replace my already 4K camera. I will not buy an F5 which is a 2K camera with raw option, that would be taking a step back and buying the exact same camera I already own for much more money. My Scarlet is much cheaper, already shoots 4K raw, and if I need to I can hook up a hyperdeck shuttle and do pro-res 422. Why would I then spend $20K on an F5 to get the same functionality as my RED camera??

The F55's advantage over the scarlet is the ability to have the internal 4K WITHOUT having to do the mess of raw for clients who really dont have the systems to handle it. 4K internal can be handled on a regular MBP with their new codec in real time with no special anything needed, even older systems can now do 4K with their new codec. That means my clients have faster turn around time. I just shot a national infomercial in 3K on my RED system and post was calling me with having to deal with REDRocket issues. It would have been easier for them to get a file that can be dropped into FCP or Avid and just spit out 4K right away.

Its not that I dont want more "latitude" its that I want more advantage to shoot then I already have. I have 4K raw what I want is 4K without the raw mess, as a lot of people do. Getting an F5 is basically the same thing as buying a scarlet but for almost $25K instead of the much lower price for the same thing RED offers and with better name.

 

[starcentral]

Why would I then spend $20K on an F5 to get the same functionality as my RED camera??

Using that logic why would you then spend $35k on the F55 when you can add a 4k recorder to your RED for ~$4k? :happy:

I just see a lot of people justifying the F55 without recorder because they think 10-bit internal will be the same as 16-bit Raw recorder on the F5. I just wonder if 10-bit will be good enough on a camera who's sensor color gamut exceeds that of film?

Too bad there was no 12-bit internal codec option.

Edit: who's sensor CFA color gamut exceeds film.

 

[bruceallen]

Raw footage does require intense CPU processing since it is not performed in-camera in real-time using a dedicated hardware processor to produce a YCrBr based image format. However I'm a little suprised to hear about disk space?!

RAW is actually more efficient than YCrBr based codecs isn't it? Which is why the FS700 can transmit 4k 12-bit raw over single 3G SDI port - how many ports do you need when using YCrBr based codecs again?

Another way of looking at it (based on 2k images):
• 16-bit RGB raw file with 3:1 compression on RED is around 200mbps
• 16-bit RGB raw file with 5:1 compression on RED is around 100mbps
• Apple ProRes 4444 is around 380mbps

The reason raw from a Bayer sensor is efficient is because it's only storing red OR green OR blue per pixel.

Your comparison is apples to oranges. That ProRes4444 file could store a fine color pattern that would be totally lost if stored in raw.

Yes, raw usually uses up less disk space than uncompressed DPX files. No, it hasn't been proven that for a given bitrate, raw will beat non-raw recording in all circumstances. For example, if there is a high-detail scene with a ton of noise and part of the in-camera processing does clever temporal denoising, then I'd guess that recording the processed file at a given bit rate will beat the raw recording, because the raw compresion won't be able to distinguish between noise and detail and will allocate bits equally, whereas the processed file will contain less noise and hence compress better.

Yes, a raw workflow on a Sony F55 would use considerably more storage space than recording 4K to XAVC... but that's because the XAVC 4K codec they're using was chosen to have a lower bitrate.

Personally I'd love it if the XAVC files had a slightly higher bitrate. The compression is at 600mbps for 60fps is pretty low but just enough to be slightly irritating in certain situations in my opinion. The compression differences will be far, far more noticeable than any 16-bit linear vs 10-bit log differences, in my opinion.

Right and shooting 12-bit is not an option on the F5/F55 which is why earlier I said people might want to reconsider when they justify buying the F55 without ext. raw recorder because it has 4k XAVC internal codec which is 10-bit. I hear it all the time lately, "I'll just buy the F55 and shoot 4k internal and won't need the AXS recorder". For less money you can shoot 16-bit raw on the F5 and end up with a lot more lattitude in post.

To each his own but I'm trying to point out some misnomers about bit-depth and raw vs log.

Let me try again - think of two audio recorders:
- Audio Recorder #1 records at 96khz, 24bits.
- Audio Recorder #2 records at 192khz, 32bits but is heavier, larger, uses more battery power and requires decoding in post. But it is cheaper.

The question is: You know that the microphone and the preamp is probably the bottleneck for both Audio Recorder #1 and #2, right?

Do those extra bits actually contain real audio data that is being lost? Or is it mainly recording noise and stuff that will never make it into the final picture?

Audio Recorder #1 may be a much better choice if the advantages in weight, power, bulk and workflow make sense.

There is a threshold of quality that is sufficient given a signal chain. And while Sony's F5 and F55 sensors looks good... I don't believe that they are so fantastic that you'll be able to pull detail out of the raw signal that isn't present in a lightly-compressed, 4K, wide-dynamic-range 10bit Log signal.

Bruce Allen
www.boacinema.com

 

[bruceallen]

Using that logic why would you then spend $35k on the F55 when you can add a 4k recorder to your RED for ~$4k? :happy:

I just see a lot of people justifying the F55 without recorder because they think 10-bit internal will be the same as 16-bit Raw recorder on the F5. I just wonder if 10-bit will be good enough on a camera who's sensor color gamut exceeds that of film?

I'm not sure that the sensor gamut does not exceed film in all directions. Only in some directions.

All we've seen is a 2D gamut plot. Don't think of gamut as a 2D plot - think of it as a 3D volume.

Film usually has much wider gamut in dark saturated colors. Wide-gamut has digital is usually only actually wider in light saturated colors.

That cute 2D gamut image Sony has posted is likely a best-case scenario.

Too bad there was no 12-bit internal codec option.

Seriously, you don't think that 73 values PER STOP is enough?

Do you really think that sensor noise isn't going to get in the way of discerning more values than that in most cases?

Bruce Allen
www.boacinema.com

 

[Kolor-Pikker]

Good points Bruce.
Now that I think about it, the dynamic range of video sensors is probably a lot like audio DR, a higher bit depth will probably only serve to quantize the noise structure than bring any actual improvements. Canon C300 does great things with just 8-bits.

In any case, we don't have monitors that resolve beyond 10-bits, render color spaces larger than Adobe RGB, or have contrast ratios that can truly reproduce 14 stops; so stop worrying too much, we're all playing with imaginary data here that's of more use creatively than technically. Technically, we're spoiled as heck, there's no saving us LOL; it's my understanding that to be a modern man is to take technology for granted the moment we find out it actually exists.