Thoughts regarding S35 vs. FF35 sensor size

[Dominic Jones]

Most of us shooting for theatrical release are going to be mastering to the DCI spec soon enough anyway and for all practical purposes and that's a 4k standard so while it'll be nice to have a 5k camera master I'm pretty happy working 4k and will be for the foreseeable future.

Hi Jonathan,

Whilst I essentially agree with all your points, I thought I'd add in a tuppenceworth of my own on the 5K front - the reason why 5K becomes a "magic" number for 4K delivery is that the bayer pattern approach yields just over 75% accuracy compared to full RGB approaches. Therefore, 5K acquisition is almost ideal for creating 4K deliverables, whereas as 4K acquisition falls a little short of the mark.

In actual fact, assuming a similar level of accuracy as the Red One (which is ~78%, iirc), you would want a minimum bayer resolution of about 5.1K to produce a "true" 4K image.

By the maths of it, assuming an S35 window size of 24x18mm, and that the FF35's active (6K) sensor area is true 135 size (36x24mm), the S35 resolution of the FF35 would be exactly 4K, so the question then becomes: Do you want oversampling to create a truly full-colour image, or are you willing to sacrifice a little colour accuracy for higher DR?

The noise issue will be interesting, as oversampling also reduces noise, so that battle can really only be fought once real-world tests can be done side-by-side. My guess would be that (unless there's a massive jump in performance as regards noise between the two) the differences will be negligible - but that's a total guess.

Of course, with most productions currently finishing at 2K (as in your experience), there's really no question that the FF35 would yield a better image.

Interesting times, and no doubt...
Dom.

 

[JonathanF]

Dom,

Your point is well taken, but as we see with Primos or MP's, the construction of Cine lenses requires they be a lot larger than still lenses for the same focal length. I don't think I want to work with optics that are that much larger (and heavier) than what we're using now just to support a larger format sensor.

I'm also pretty sure that MP quality glass at this image circle size will require selling children into slavery or other equally distasteful options. I don't hear any discussion from manufacturers about new ranges of lenses either. Sure, you hear the odd rumor now and then, but never anything concrete (no offense to anyone but I'm not talking about Sygma glass or Nikon SLR lenses here but real top tier Cine lenses).

In this economy I don't see any of that happening (or the producer's paying for the increased rental on the lens package for an 8 week shoot). I know I've been in meetings where 65 was raised and it was always a really short discussion. Costs go up exponentially on all fronts. The budgets just aren't there these days to justify it. Go ask ILM what the price difference is for VFX at 5k vs 4k, or 4k vs. 2k. The discussion is always real short when production gets the numbers.

As a bit of a digression from the lens issue, I would also encourage you to go look at the difference between images produced at lower bit depths with more resolution vs those at higher bit depths at slightly lower resolutions.

If you can find a suitable monitor like one of the 10 bit HP Dreamcolor monitors the facilities are using as reference monitors now for VFX and compare the same frame at 4k 8bit (what you get to graphics on all but the most recent Nvidia 10 bit cards released earlier this year) and compare to 3k at 10 bit you'll be pretty surprised.

Put all this cine stuff aside for a moment as you can do this test with two HP monitors, one 8 bit and one 10 bit (assuming suitable graphics cards as the 10 bit DisplayPort interface graphics cards are exceptionally recent) and a high end DSLR with Raw source in photoshop. Remember DVI-D is only 8 bits to graphics. You'll need to use the displayport interface to actually get the 10 bits to the monitor.

Mind you you'll need to calibrate the monitors so they are comparable in terms of white and black point, but there is an astonishing difference.

I have to say I see dramatically more improvement in the perceptual image quality when I get more bits then I do when I get marginally more resolution, and at 24 fps the perceptual difference of more resolution to the viewer at 50ft of distance is going to be negligible compared to a higher bit depth master.

But the big issue that hasn't even been broached is the fact that there are only two common projection formats or aspect ratios, 1.85:1 and 2.39:1. I know Vittorio (for whom I have immeasurable respect) has shot some 2.0:1 (Univisium) stuff that was released in the original format, but most of us don't have the moxie to pull that off in a meeting with distributors even if we can get it past a producer.

The simple fact of the matter is, the vast majority of all theatres are set up and have lenses, aperture plates, and screen masking for just two aspect ratios, 1.85:1 and 2.39:1. A small percentage of theatres (although not in the majority) are set up for 1.37 or 1.66 but they are a less than single digit minority, and are generally boutique venues.

From a practical business standpoint releasing in anything other than 1.85 or 1.39 is going to cost you so many screens the distributor is going to crush the idea before the ink is put to paper on the contracts.

With that in mind; the 2.0:1 sensor shooting for 1.85:1 theatrical release tosses 7.5% of the image away on the sides with the 2.0:1 sensor.

If you're using the 24x36mm sensor you only use 81% of the sensor giving away 9.5% on both the top and bottom of the frame.

I won't do the math on the scope relationship for this discussion as I'd have to go looking at what scope optics are even available for a sensor of that size before I even think about it.

So how the bayer filter matches up to what resolution is a little more involved than just saying 5.1k is actually 4k, because you have to know what format your release will be projected in because that's the only relevant portion of the horizontal dimension of the sensor at the end of the day.

From a practical, in the field, solving every day production issues standpoint, I'll take the extra stops over an extra 1k of resolution.

I do however appreciate the technical thoughts on the bayer resolution issues, but there are the practical mechanics of getting films into theatres that need to be kept in mind if this discussion is to have any relevance in the real world. I just don't want people thinking it's as simple as more resolution is better. That's the kind of logic you hear in the consumer market, and it's really more complicated than that.

More bits for me please.

Cheers,

J

 

[JonathanF]

Dom,

Another quick note on the noise issue is that while resolution does improve the noise issue, those who have shot with eng cameras will note that gain has an equal if not greater effect on the noise and that never sensors will likely improve on the sensitivity of earlier sensors.

This alone will likely result in a greater reduction in noise than just adding resolution.

I have yet to see any real discussion on this, or how the new sensors differ from the R1 sensor in this area, but if you look at the image in each of it's component channels you'll see clearly that the RedOne, like other CMOS and CCD cameras, requires a bit of additional gain on the blue to compensate for the narrower wavelength (tapping into 20 year vintage physics from college here so permit me a little leeway).

This is why some people will use 80 series filters to improve the quality of green screen plates and using tungsten lights (or compensating for the color temp shift in post). It makes the guys who have to pull keys in post's life a lot easier as the reduction in the noise is pretty drastic. You just add that 2/3 stop in up front when you shoot and Bob's your uncle. It's low tech noise reduction but it works like a charm.

Cheers,

J

 

[Dominic Jones]

Hi Jonathan,

Just wanted to follow up a couple of points here - I'm not sure where the discussion of lenses fits in, as we're talking standard S35 "gates" on both systems, so I'll skip that part, unless I've missed something?!

There's no doubt that higher bit-depth is always preferable all other things being equal, but the equation between bit-depth and "true" resolution still stands, and it is one that each DP will have to decide upon (often for each project) - of course, this thread relates primarily to those who are going to purchase, which means people have to decide which direction will be preferable for them (or their rental clientele) in the majority of situations. Just to be clear, I am not "supporting" either point of view (I know my preferences, but this isn't about what I would or will do) - rather, just trying to clear up the seemingly myriad choices for those who prefer S35-based shooting, as you yourself seem to (and I'm with you there for the majority of cases!). It is also worth noting that differences in ratio are irrelevant to this conversation, as both are wider than the sensor's native 1.78:1 ratio, and we're working with horizontal resolutions only, which are independent of ratio, of course.

Also, please bear in mind that I'm talking about the specific case when 4K delivery is required. I am fully aware that this is - currently - rarely done and often cost-prohibitive, but anyone who expects that to be the case for long will be sadly mistaken - Moore's law is in effect now with a fully digital path through principal photography, post and on to distribution, and many of those planning camera purchases now will be wanting to ensure that their particular bird still flies as and when 4K becomes a larger part of the distribution market...

Regarding noise and exposure, you've kind of answered your own point! if you're going to add an extra 2/3 stop (or 1, or 1+1/2, depending on filter of choice - 1+1/2 (an 80B) being ideal for an R1), then you can equally add more light to compensate for a less sensitive sensor!! In a documentary environment you do make a valid point, but for drama and promos my post assumed a well-exposed image shot at native sensitivity - apologies if I didn't make that clear... Of course, there are ancillary benefits to faster sensors (such as smaller and less costly lighting packages), but let's stick to the camera sensitivity side of things for the sake of simplicity!

Hope that helps to clear a few points up, I think essentially we're both on the same page!!

Cheers,
Dom.

 

[JonathanF]

as both are wider than the sensor's native 1.78:1 ratio, and we're working with horizontal resolutions only, which are independent of ratio, of course.

Actually on the S35 sensor native is 2:1. My preference would be an academy formatted sensor to be honest. That's the only thing I can really think of that makes the Arri even worth consideration. Makes shooting scope a bit more attractive, although arguments can be made that even cropping the sensor on the S35 X will be better, so we'll have to wait and see how that all plays out (and how we get to view scope in the EVF and LCD).

Hope that helps to clear a few points up, I think essentially we're both on the same page!!

Yeah, I agree that we're pretty close here, but I really have to say the whole resolution thing is a bit distracting from the real discussion we should be having which is how do we actually get a better image.


J

 

[Dominic Jones]

Ah, ok - didn't realise the S35 will have a 2:1 sensor - that does change the issue a bit... Are you taking that from the stated 30x15mm sensor spec (which almost undoubtedly includes blanking pixels and the like), or is that a confirmed spec from Red for the recording area? Would be a very odd choice for an "S35" sensor if it is the case...

And I totally agree - image quality is the key. As I said, I'm thinking for these posts not so much in terms of what I want from the system, but rather the issues others are talking about, particularly as regards 4K delivery.

There's no doubt, in my mind at least, that less noise and higher DR are much more beneficial to the final image than huge resolution, but with an eye to the bayer downsampling that is so very important - having shot 2K material for a 2K finish with the R1, I can safely say it is not really up to par compared to 3K/4K material for same, so assuming that a similar situation is true of the Epic's sensor, I would definitely want >4K resolution for a 4K finish...

I'm more than happy to be proven wrong by the Epic's images, though!

Cheers,
Dom.

 

[JonathanF]

Ah, ok - didn't realise the S35 will have a 2:1 sensor - that does change the issue a bit... Are you taking that from the stated 30x15mm sensor spec (which almost undoubtedly includes blanking pixels and the like), or is that a confirmed spec from Red for the recording area? Would be a very odd choice for an "S35" sensor if it is the case...

May I point you to: http://red.cachefly.net/14/mysterium.jpg

That has the sensor dimensions. I am in 100% agreement with you on the sensor size, I'd much rather not see a 2:1 sensor, but that's what we have in the R1 and it looks like that's what we'll be living with on the Epic S35. In a perfect world I'd like a 0.980 x 0.735" sensor. That would make every existing (quality) cine lens out there usable. Especially the good scope glass. I'd be slapping a PV mount on for sure then, especially if the viewing issues with scope were resolved.

so assuming that a similar situation is true of the Epic's sensor, I would definitely want >4K resolution for a 4K finish...

...but I'm going to reserve judgment on that till I see what redcode 250 looks like. That could easily be a game changer.

As for the larger sensor, I'm sure it will look good but it's already hard enough to pull focus in scope with a S35 frame, I don't even want to imagine what my focus puller is going to say if he has to pull for scope on a FF35 frame. Can you imagine a 100 at FF35? at f4 you'd have like a cm of leeway.

Wide open you'd be screwed.



J

 

[Dominic Jones]

The sensor in the R1 is not 2:1 - it is 1.78:1. There is a 2:1 recording mode (and this was the first mode enabled), but the sensor is 1.78:1 and can be recorded as such...

That said, the S35 chip does seem to be wider in ratio, but we'll have to see what the specs released end up being before jumping to conclusions, I think.

I do agree that an Academy ratio sensor would be great for anamorphic shooting, but I imagine the thinking is that many people would not want to pay extra for all that sensor they're not using much - not that many people shoot with anamorphics as compared to spherical lenses in the Red user-base, as far as I can tell (I would imagine that the ratio is very different for Arri, however).

Yes, RC250 may make a big difference, although my gut instinct is that the issue is sensor noise, not compression artifacting - although of course, with a new sensor that may all change as well... I also doubt whether - short term at least - many people will be recording using the 250MB/s (or 225MB/s) modes very often. That's a hellish data rate to record real-time.

As for FF35 focus pulling, yes it will be more of a challenge, but not insurmountable for a good AC - and remember, a 100mm lens on an FF35 is exactly the same as a 100mm on an S35 chip, except for the FoV - DoF does not change with gate size (although your choice of lenses will become slightly longer). I have personally pulled on very long lenses wide open, and it's not that big of a deal, if you've got experience. Macro work is by far the most challenging, often with a DoF measured in millimeters! Of course, if you're planning on focusing yourself while operating, by eye, then that's a different issue!!

 

[JonathanF]

The sensor in the R1 is not 2:1 - it is 1.78:1. There is a 2:1 recording mode (and this was the first mode enabled), but the sensor is 1.78:1 and can be recorded as such...

Oops, you're right Dom.. my bad. I need some more sleep.

I have personally pulled on very long lenses wide open, and it's not that big of a deal, if you've got experience. Macro work is by far the most challenging, often with a DoF measured in millimeters! Of course, if you're planning on focusing yourself while operating, by eye, then that's a different issue!!

Well, I like to think I've got a pretty talented focus puller and with the 150 MP wide open at 1.3 from close up it can get real challenging, and I've been doing it for more than 15 years. I like to be a little kinder and give credit where credit is due in a situation like this because pulling focus in this shot (Check your PM for a couple links) was _not_ easy.

But you're right on the R1 sensor. I've been in the studio too many hours this week. I'm not looking at shooting scope with the R1 because of the viewing issues in the EVF and the LCD, but if that ever changes, or if the Epic resolves those issues, I'm very keen on doing my next scope film digitally on one or the other if I can slap a PV mount on the camera. Then the focus will get real interesting.

 

[Dominic Jones]

Yeah, no doubt that a 150 @ T1.3 ain't so much fun!! I'm probably a little spoilt these days as my FP is a ninja! And all credit where credit is due, it's a very tough job (I know, I've been one!!) at the best of times - I just don't think that FF35 is going to be such a total game changer (I probably won't be shooting much of it anyway though!). I also doubt you're going to see much T1.3 glass that covers the Vista frame any time soon...

Now get some sleep!!

Cheers mate,
Dom.

 

[Dominic Jones]

Oh, btw - Build 18 for the R1 offers de-squeezed anamorphic monitoring, if that's of any interest to you?

Just in case you hadn't noticed - it can be a full-time job keeping up with the Red code dept!

 

[JonathanF]

Dom,

I have at least one more drink in me before I fall down.

And yes, that build 18 DOES interest me if that's in there... now I'm going to have to play with it... you just made my week off less relaxing.

Thanks mate!

J

 

[Dominic Jones]

Ha ha, sorry bro!!

Does indeed include anamorphic de-squeeze, but do bear in mind that it's beta and has been for some time. I've tested it (moderately) heavily, and it seems stable, but have not (as yet) shot in anger with it - others may well have, so perhaps try (a) searching for Build 18 threads and (b) posting a separate thread under the R1 forum - we're pretty well OT here!

Off for a wee dram myself - I'll have one for you!!

Take care mate,
Dom.

 

[david farland]

Nice thread....figures published in Dec (& few more) for sensors were....

R1 = 5.4 micron - 4K - 24.4 x 13.7mm - 28mm (diag)- 10+ stops - 4K (horiz) - 3.2K (luma rez) - 2.7K (chroma rez)
Mx = 5.4 micron - 5K - 30 x 15 mm - 33.5mm (diag) - 11+ stops - 5K (horiz) - 4K (luma rez) - 3.4K (chroma rez)
Mo = 6.0 micron - 6K - 36 x 24 mm - 43.3mm (diag) - 13+ stops - 6K (horiz) - 4.8K (luma rez) - 4K (chroma rez)

The major points of 6K over S35 will be 13 stops (using larger 6 micron pixels) & 4K chrominance but you'll need the FF35 lenses (hello Viewfactor).

Pawel tested his master primes (here). 14mm & 35mm lens had a 31mm image circle. The 6K sensor has a 43mm image circle (max diag).
I guess people will buy several mounts and use whatever lenses they have or can afford to get full 6K.

Dave,

 

[JonathanF]

R1 = 5.4 micron - 4K - 24.4 x 13.7mm - 28mm (diag)- 10+ stops - 4K (horiz) - 3.2K (luma rez) - 2.7K (chroma rez)
Mx = 5.4 micron - 5K - 30 x 15 mm - 33.5mm (diag) - 11+ stops - 5K (horiz) - 4K (luma rez) - 3.4K (chroma rez)
Mo = 6.0 micron - 6K - 36 x 24 mm - 43.3mm (diag) - 13+ stops - 6K (horiz) - 4.8K (luma rez) - 4K (chroma rez)

So in theory then you should be able to do partial sensor extraction on the Ro and that could for all practical purposes result in the ability to shoot with a 23.76 x 17.82 sensor at 3.96k or 3.168K luma rez and 2.64 chroma rez. You would then have the same sensor dimension as the D21 and 13 stops instead of 11.

That kind of settles it for me.

Now what I would want in that case is a shooting mode for partial sensor extraction with just a little 'look around' room that's not going to disk. Seems that should be possible.

For a $5k premium it seems a no brainer to me.

Thanks for the post David, I really hadn't thought about it in these terms.

:J