Scarlet 2 - Bored 2.0 ;)

[Wayne Morellini]

13-09-11
Update on recent teasers from Jim and speculation on Scarlet and Epic-S changes, and possible update of model specifications.

Suggestion: 02-08-11 s16-4/3rds, 8mp+, higher dynamic range and sensitivity.

See second post for update list.


Firstly, unfortunately this light hearted thread comes after too many complaints expecting red to make changes when it's too late. So, lets just have fun with what we would like in the future (realistically).

This mainly about the bottom end scarlet option for every professional below a cinematographer.

It is getting late in the year. I prefer a 4k camera as a standard for future work. There is possibility of 4/3rds gh2 or 3 and f pro series hack to do this, but I prefer Red level of quality. So maybe I will wait for the next scarlet.

What do we know about the next iteration of sensor technology, Monstro?

I would totally prefer 4/3rds 4k, even with 8k stills, or 4k 2/3rds. 2/3rds lens could be used on 4/3rds, and 4/3rds lens for photography (or video). If a 2/3rd, 4k resolution, it would need a smaller lower performance pixel, but scarlet is the lower end camera under the Epic anyway.

Cheap single lens 3d system.

Ambarella makes cheap powerful android based camera chip set systems suitable to be interfaced to a Red's sensor technology and probably powerful enough to compress redcode. Such a system in a consumer alloy enclosure, could be done for similar cost to the gh3 plus the red sensor and maybe compact stock full manual and auto red full zoom lens for basic video and stills. It could be offered jointly through a still camera company as an top end camera. Olympus and Sigma are both in need of better top end cameras, Sigma also having compression friendly 3 color sensor technology for partnership. This would replace the fixed and maybe more.

To give an idea of last year's android chipset, it was reported it could do 5mp video. The only issue is either getting it to compress redcode, or doing it externally.

The Question:

What would we need for a basic camera like this to film truely cheap indie film, documentaries, commercials, weddings, training and whatever videos, photography and video journalism, where the majority of non consumer sales are. Also, what sort of lens would cover this range of work.

Anyway, have fun, what would you guys want?

I would like to offer this thread as a standard place to discuss the real next Scarlet 2 base model, and possible suggestions for it. Also, offer it as a possible sticky, if people can summarise their suggestions and information for possible inclusion in the first posts.

Re-Edit: I forgot, I am naming this camera suggestion Red - Violet.

 

[Wayne Morellini]

13-09-11
Update on recent teasers from Jim and speculation on Scarlet and Epic-S changes, and possible update of model specifications.

02-08-11
After much irrelevant off topic debate ; )
s16-4/3rds, 8mp+, higher dynamic range and sensitivity.

31-07-11 (my time zone)
Suggested that super 16mm version could keep some pixel size with 4k. (by David).
Good suggestion, do we know for certain.

 

[Wayne Morellini]

Place holder2.

 

[David Rasberry]

The 2/3"/16mm sized format has pretty well proven itself over decades to be a practical sweet spot for general purpose production work.
I like M4/3 as a format but it is still in its infancy. So far it is in an optical no man's land where one is pretty well limited to using 35mm format lenses. No one has stepped up to produce the kind of fast wide range EFP zoom lenses that are the bread and butter of the 2/3" format. Such lenses would be twice the size and weight of 2/3" lenses for equal optical performance too.

I'd like to see Scarlet interchangeable morph into a true S16mm sized QHD 3840 x 2160 sensor, something which could be accomplished on the existing 3.3 micron pixel sensor design of the 3k 2/3" version. But I don't think any off the shelf aftermarket chips could handle Redcode efficiently or cheaply even at Scarlet levels and this would require near Red 1 processing power to match the 2/3" performance features.

Otis Grapsas has apparently modified his Drama camera software to work with several off the shelf 2k industrial cameras recording uncompressed to a laptop, but his full self contained version still has a target price around $10k with all off the shelf components, so sub $5k cheap raw motion camera is still not a realistic expectation apparently, even at 2k/HD resolution. Existing industrial camera heads without on board processing of any kind go for between $2k and $4k at full motion HD resolutions

 

[Dan Hudgins]

Its possible to make a 2.5K RAW recording "web cam" like buffered camera for about $250 using maybe 24 CF cards for the buffer. When you get a internal PC in the camera the weight and cost and power use go up too much.

Since I am just interested in feature filmmaking a buffer time of between 30 and 60 seconds is enough, and shooting only 24fps is fine. Most feature films made (to date) do not have any shots longer than 60 seconds and most only have 24fps footage.

So you have a buffered head with USB or LAN to a netbook for the viewfinder, small netbooks cost less than $300, you then plug in a usb external 2TB HDD enclosure and you can shoot up to 2TB by using the "shoot-then-dump" system.

Its the highest quality result for "indie feature" shooting.

You can use the HDD enclosure, netbook, and CF cards for other uses or sell them on ebay when the film is done, so the only item that would be a non-standard computer part is the camera head with the $50 2.5K sensor and OLPF+IR/UV cut filter.

I would not have any issues with shoot then dump, as the wait between takes would not be a major problem if the CF card module could be swapped for making re-shoots with less delay, then dump after all your memory is full, that is called "shoot-then-swap".

The same camera design can be used for "shoot-then-dump" and "shoot-then-swap".

With Scarlet RED has tried to make something for the market that gives many features that are not needed for "indie feature" shooting since their target market for Scarlet is not "indie feature" shooting as such is not profitable, people who are going to fork over $8000 to $12000 for a Scarlet system and lenses and accessories are going to need to be professional or well to do, and if Professional they will be shooting things that can give a return to pay off the not-so-cheep high quality Scarlet camera system.

I don't think RED would lose any Scarlet sales if they were to offer a high quality $250 true RAW recording camera as it is a different target market, one inhabited just by filmmakers that cannot to spend more but want it to look as if they had and are willing to deal with some time vs. money trade-offs.

 

[David Rasberry]

From a production viewpoint, that all sounds very inconvenient and inefficient, even for a no budget indie. There are multiple high quality industrial camera heads out there with a variety of sensors that will dump raw high quality Bayer data at motion rates, but they are nowhere near $250 cost. You still have to get it into some kind of workable form factor for a camera op. Mounting systems, lenses, viewfinders, monitors, practical recording solutions. Being tethered to anything external is not very practical outside of a studio.
Even as a serious hobbyist I am not willing to be hobbled by impractical shooting requirements just to get raw output. The aggravation would far outweigh the cheap price. I would rather use a low cost HD video camera or DSLR in that case. Even DSLR's need significant hardware added to be practical motion cameras.
If you are talking about any kind of long form feature film, other production costs are going to far outweigh the camera cost for any film of high enough production value to be reasonably marketable. Inefficiency can be very expensive.

 

[Wayne Morellini]

The 2/3"/16mm sized format has pretty well proven itself over decades to be a practical sweet spot for general purpose production work.
I like M4/3 as a format but it is still in its infancy. So far it is in an optical no man's land where one is pretty well limited to using 35mm format lenses. No one has stepped up to produce the kind of fast wide range EFP zoom lenses that are the bread and butter of the 2/3" format. Such lenses would be twice the size and weight of 2/3" lenses for equal optical performance too.

I'd like to see Scarlet interchangeable morph into a true S16mm sized QHD 3840 x 2160 sensor, something which could be accomplished on the existing 3.3 micron pixel sensor design of the 3k 2/3" version. But I don't think any off the shelf aftermarket chips could handle Redcode efficiently or cheaply even at Scarlet levels and this would require near Red 1 processing power to match the 2/3" performance features.

Otis Grapsas has apparently modified his Drama camera software to work with several off the shelf 2k industrial cameras recording uncompressed to a laptop, but his full self contained version still has a target price around $10k with all off the shelf components, so sub $5k cheap raw motion camera is still not a realistic expectation apparently, even at 2k/HD resolution. Existing industrial camera heads without on board processing of any kind go for between $2k and $4k at full motion HD resolutions

Thanks for the s16 suggestion David, I have listed it as the first update on the second post. People can just check the first posts to get upto speed and for the latest update notice.

On the suggestion, that is a really great idea, but I do not know the actual measurements are. Does the pixel size pretty much exactly match for the sensor aspect ratio when you divide 4k into the super 16 frame?
People get hung up on the 4/3rds sensor idea, but like the s16 sensor, you can mount a 2/3rd lens to it (even cropped 35 if you really wanted) and loose nothing but a little resolution, but still 2k and more than many 2/3rd lens can resolve. You also get to use still 4/3rds for photography. So it is a more or less, relatively, no loss situation.

As for the 4/3rds lens situation, eventually a compact lens could be done that people could use, and there is nothing stopping Red from taking to it thenselves, if they wanted to, I'm sure other 4/3rds owners would be interested.

The ambarella crowd come from industry leading image processing experience, it is remarkable what they achieve compared to the competition. The simplicity of redcode may help take advantage of the increased parallel elements of the latest design.

You can get an Elphel raw uncompressed recording head with sensor for less than $1000, then you may still need a cheap device to preview, and I do not know about 2/3rd or larger sensor options. The cinema camera setup is still under development, so you would have to dig a bit to find these details. Sumix project at dvinfo, the granddaddy before the Red project, used an advance but relatively cheap 2/3rds 720p compatible sensor. Sumix now offer a 2/3rds 1080p sensor like that used in the si camera, and guys at dvinfo are developing digital cinema camera software for it, nothing yo be sneezed at. So, a system with small computer ( not forgetting other available hardware recorders) could be done for less than $2000, even less again. But a consumer camera with manufacturing volume that had 4k could be done for $200-300 plus storage and sensor. I say 'plus' because getting a good 2/3rds video capable FHD+ sensor canbe reasonably expensive, unless you can order vast amounts of them. Nothing is as it seems, volumes speaks volumes of cost reductions, but thus economic climate foes nit speak volumes of sales to get there.

 

[Wayne Morellini]

Firstly, Dan, that webcam thing sounds interesting, are you referring to an existing project?

David and Dan, but what do we need in a base model second series Scarlet that is specifically not for a major feature or indie film, but for everybody else?

Thanks again.

 

[Wayne Morellini]

Ok, looked around for the Reds Dragon sensor technology, and is for late 2012 time frame. From this we can conclude that though a refreshed Scarlet sensor might be possible beforehand, it is unlikely. Sensor performance normally improves as their manufacture improves, so latter scarlets could still possibly indirectly benefit a little.

Re-edit: Ok Dragon is a particular sensor or sensor upgrade. Monstro is the next sensor series.

 

[Dan Hudgins]

True RAW vs. something else

True RAW vs. something else

What "Elphel raw uncompressed" are you talking about?

This link shows JP4,

http://cinema.elphel.com/elphelcamera

Do you have a link for "actual" 12 bits ADC sensor output being recorded from something manufactured by Elphel. I went round-and-round with them before about adding a simple 12bit port to there camera so that an external True RAW buffer to USB could be added and that meet a stone wall.

The grading range of 8 bit data with out adding histogram gaps is nil.

The bandwidth for their 2592x1120x1.5x24 is 104.5MB/s which should be too large for USB and LAN, so that would leave their SATA port, have the actually recorded real uncompressed 12bit sensor data (as it comes out of the ADC) without errors or glitches yet? Or do they use curve encoded data recorded as 8 bits?

==

I have been developing de-Bayer software for actual uncompressed Digital Cinema Cameras for the last several years, the code is in final alpha development and has been tested along the way with actual True RAW sensor data from various cameras under development.

I original approached several camera makers before RED came out with a camera about their so-called 24fps board cameras, it turns out that they seem to fib about those camera's actually being usable at 24fps in this way:

1) The sensors can shoot up to 30fps.

2) The USB cannot pass the sensor bandwidth glitch free at 24fps in real operation.

3) Laptop computers do not have RAID SSD needed to store the data glitch free at 24fps.

I asked one of the camera makers if I could come over with a clapper-board slate and shoot some sync sound tests since he should be able to demo the 24fps recording and he then did not respond to my emails.

I asked Sumix what computer system requirements were to ACTUALLY record at 24fps sustained, and only got an answer back that they were something like investigating the matter. Someone saying that they were a user sent me some processed data that had undergone gamma correction, noise reduction and poor white balance and said he could not record higher than 960x540 on his computer. When I asked me to send me some actual RAW data that had not been mucked up he did not reply, so I asked Sumix and they also did not respond to my requests for reference RAW data to establish the EI/ISO of their cameras. I added support for their 12bit format as best I could. Their camera does not seem to have a OLPF filter, making it for the most part useless in filmmaking due to unfilmlike aliasing, perhaps. You can add one I guess, but its just silly to expect the camera buyer to make complex mechanical modifications of their cameras.

==

After talking to many camera makers I determined that the cheep to develop camera they have gone for are not the cheep to purchase and use cameras that filmmakers need.

I approached a camera maker about adding a DRAM buffer to one of his low cost cameras, and he said that he could add a FLASH buffer, but rather than doing that he ended up making another "computer cam" like NOX-2K.

Computer cams have the problem of glitches from the OS that require a large FIFO between the sensor and the SSD to record to.

==

Someone that knows people at ACAM (he said) contacted me (why I am not quite sure) and I told him that the lowest cost Digital Cinema Camera with True RAW recording can be made using a sensor and buffer, since it seems that there are no people around the world that understand how to make a DRAM buffer, using CF cards is possible because their interface is easy to understand (RED already understands how to write to CF cards).

A buffered camera has these advantages:

1) The speed of the computer does not matter as far as data glitches, any low cost Linux net-book from China can be used as the viewfinder and pass through to save the data to an exteranl HDD who's speed does not matter as well.

2) The bottle-neck of the USB or LAN are no longer any issue as far as the shooting frame rate or data glitches in the recorded data, you get the exact frame rate you want, zero dropped frames, and no glitches ever.

3) If the memory module is designed to be swappable you can shoot-then-swap as well as shoot-then-dump for more flexibility.

4) Since USB/LAN bandwidth do not matter longer cables can be used between the camera head and the computer.

The primary objection seems to be that the buffer does not allow for "video like" continuous recording. Continuous recording is not required for feature filmmaking, a 30 second to 60 second buffer is all that is needed. If you wanted to spend unnecessary extra money you can make the buffer equal the shooting time of a 400' or 1000' 35mm magazine and if someone argues that that is not enough for feature filmmaking then you should question their knowledge of filmmaking history.

==

As to the remarks about ease of use, yes a self contained camera is easier to use, I cannot comment at this time about progress on their development.

I should like to point out that if I remember right SI-2K mini were used on "Slum-dog" plugged into notebook computers recording Cineform (tm) compressed data (non white balanced wavelet similar in many ways to REDCODE).

==

Not having a compression engine in the buffered "web cam" camera has these advantages:

1) Dark field calibration can be done both before and AFTER the shots have been made.

2) The size and cost are MUCH lower as no CPU or GPU are required in the camera as no processing of the data is done or needed.

3) The recorded data has not burned in dark field calibration issues like REDCODE (tm) or Cineform (tm), nor does it have any wavelet blur or chroma or luma artifacts caused by wavelet compression.

==

If a True RAW camera is lower cost and somewhat able to produce image with fewer artifacts, then where does the opposition come from with regard to making such an addition to the other RED cameras?

The main arguments seem to be that you cannot shoot slow motion with such a buffered camera because the buffer is not large or fast enough and ALL RED cameras MUST shoot slow motion at high frame rates. Well, if slow motion is the prime engineering requirement for ALL RED cameras, then maybe someone else cam make a 24fps camera that works and costs less, and RED should not care since it does not compete with their compressed high speed cameras, right?

==

The software I have developed (hopefully to be released soon when some more issues are resolved with camera makers) is a "universal" system that can process True RAW sensor data, as well as DNG and some proprietary data such as Sumix (tm). (For REDCODE or Cineform (tm) cameras it processed DPX 10/16bpc or TIF 16bpc from their de-Bayer for CC etc.)

Delta-encoding can be used to encode RAW data to reduce bandwidth on average and increase recording time and is 100% lossless and can easily be proven to be, its like ZIP in that it is 100% reversible. Its not needed as it is possible to build the low cost buffered camera without it.

Using un-buffered "industrial" cameras may have these problems:

1) High moving pattern noise from poor power supply filtering and other layout issues.

2) Lack of OLPF as a standard item.

3) Requires a very fast computer to record onto to, and so draws much more power than using a buffered-camera because real-time high bandwidth is needed OUTSIDE the camera body. The very fast computer is also much heaver and bulkier.

4) Not designed with cost optimized parts and offered at small margins to fit the target filmmaker's budget.

5) Cannot work at the stated and needed fps and resolution without data glitches.

==

The best buffer would be DRAM for "shoot-then-dump" cameras, since DRAM does not wear out with repeated uses and does not need a long erase cycle.

==

RED has good sensors, it would be very simple and low cost project to offer a buffered USB/LAN camera without compression.

BTW, there is no reason that a buffered camera head cannot be used untethered if you hook a battery to it and use an optical side finder, or include some simple viewfinder circuit for off tether use.

 

[David Rasberry]

Red Dragon is probably the first generation of Red's full frame 35mm sensors, larger pixels which means lower specific resolution.

Many shooters on this site that do commercial/industrial work would really like to see an option to record 1080p Prores files in camera like the Arri Alexa for fast turn work. That is one of the single most requested features. It would make a lot of practical sense for Scarlet too.

Talking about a cheap mass produced 4k raw motion camera is fine, but the cost of hardware is really a very small fraction of the cost of a production model camera. It is the engineering, research, development time for both hardware and software that ultimately determines price in relation to production volume. If what Red is doing is so easy and cheap to do, why hasn't anyone else done it?
Red is pretty much packing miniaturized super computers into Scarlet and Epic. Scarlet at 120fps can process almost 600 million 16bit pixels/second, Epic more than 1.6 billion in a package about the size of a Hasselblad. Not just image recording, but supporting multiple independently configurable monitor outputs, a comprehensive range of color and exposure metadata, automated lens controls, file compression, audio, time code, genlock, GPIO controls, USB, Ethernet, wireless, etc. Then you have to consider what kind of manpower it takes to provide support for your intended customer.

From all the various questions and issues that come up on film maker forums regarding far simpler less demanding camera systems, I would say that Red made a smart choice to stay out of the consumer mass market and gear their company to support the more serious visual artists working in still and motion production. These are cameras best suited to dedicated cinematographers and photographers, not the low end indie/hobby/student film maker for whom finely honed skilled control over every aspect of image quality is not a major consideration. There are already many cheap and practical options to cover that market.

The typical low end indie shooter has problems dealing with cameras that pretty much offer the full range of consumer friendly features: auto focus, auto white balance, various kinds of auto exposure, predetermined optimum gamma curves, etc. They are for the most part interested in "telling a story" not in the intricate technical details of how best to visually accomplish that. They want to push a button and have the image magically conform to their imagination with as little effort as possible. How are they going to deal with a box like you suggest that is nothing but a sensor and a bit pipe?

Fact is, with all of the features and capabilities Red is building into these second gen cameras, including the modularity that lets you build a camera system in stages from very basic to very complex, there is very little I might ask for that is not already covered in some fashion. Cheaper would be nice, but these cameras already break historic price/performance barriers by a healthy margin.

 

[Dan Hudgins]

How are they going to deal with a box like you suggest that is nothing but a sensor and a bit pipe?.

I agree that RED will probably never make an uncompressed low cost camera as that was not their goal from the start. I was just saying that they easily could if they wanted to.

Since you asked the question, I have already solved many if not most of the problems you describe for the most part in the software I have written and its being tested now. I assume that manual exposure is a goal of the cinematographer, but in the system I came up with there is a "fool proof" exposure aid that relates to the meta-data and look files that allows a True RAW camera like you describe to get color matching in the BMP/CIN/DPX/TIF files without complex circuits in the camera, if the viewer in the net-book uses some simple LUT in its low res live view, the LUT are programed in a "round robin" between the post software and camera support software. It could be possible to expand on the "fool proofness" and have the viewfinder program (currently not my code, but owned by one of the camera makers) tell the cameraman which way to turn the iris ring, auto adjust the shutter angle, or tell them what ND or 80 series filters to put in the Matte Box, but I think anyone that SHOULD be using the camera should have enough skill to do those adjustments by looking at the exposure aid, but if not then its just a few lines of code to put a exposure "wizard" into the viewfinder program that runs on the net-book to aid people that need it, perhaps.


I certainly don't have a "totally fool proof system", but at least there seems to be a way to do it without a "super computer" in the camera, and that keeps the camera cost down and the image quality up. There is no "one size fits all" that is the optimum for everyone and paying for more features than one needs just adds cost. Judging by the number of people searching for help with current RED camera issues, maybe something simpler might help more than hurt the confusion issues.

I actually designed my de-Bayer program to have a configuration file and batch processing so that if a user get stuck, I might be able to configure the program for their footage and email back the "auto-pilot" files and run their frame processing that way, rather than spend time expecting them to find the right settings by themselves, they could then see what changes I made by looking through the programs menus to learn what was needed, if they wanted to. I could even grade their shots, and let them use their own render farm to do the finish processing, not that I plan on grading everyone projects for them...

==

The market for compressed camcorders will always be much larger than the market for uncompressed Digital Cinema Cameras for making theatrical motion pictures. One can argue that RED made the right choice in going for a compressed camcorder that has features of a Digital Cinema Camera, and ARRI seems to have gone even more camcorder with their white balanced prores giving up in camera "raw" recording and the subsequent processing. The market for cameras is mostly non theatrical, maybe by a ratio of 100:1. So in talking about a low cost camera intended for theatrical use, in a sub-segment of the theatrical field, it seems like a niche, and indeed it would be, but many things that are useful to some are ignored by the many.

 

[Wayne Morellini]

IThe market for cameras is mostly non theatrical, maybe by a ratio of 100:1.

And that is what I was also talking about with Violet David.

I had a look at your sig, interesting. Have you considered looking at ambarella chipsets, Aptina has development systems. www.greenarraychips.com was something I also looked into for a recorder control chips, but unfortunately wont handle a 4k stream, though maybe 2k.

Thanks.

 

[Wayne Morellini]

Talking about a cheap mass produced 4k raw motion camera is fine, but the cost of hardware is really a very small fraction of the cost of a production model camera. It is the engineering, research, development time for both hardware and software that ultimately determines price in relation to production volume. If what Red is doing is so easy and cheap to do, why hasn't anyone else done it?

It is included, what you describe is a larger part of a small run products, lesser of a mass run product. I am talking about using pre-existing mass produced parts etc. The way you describe reds features, sounds like how you could describe certain ambarella Ione chipset, and those people really deliver much higher ratios of price performance, they set them. What red is trying to do is remarkable, but also one of the most expensive ways to design a small run product, it is best you do for large run products though. In the real industry, silicon chips, they have long given up talkng about economies of scale at 100,000 or one million, now it goes into tens or hundred million. So custom silicon is not a lightly costed decision. We had people years ago going on how they were going to design a camera FPGA from virtually no experience, something a hundreds as complex as what Red is doing, with no backing. They were warned about how overly complex it would be for them, and they disappeared, hopefully not into Red (though that would explain a few things ). So a back down on the low end of the market is a shock, I would have advised against custom silicon and even potentially expensive FPGAs, for smaller runs.

So back to what people would like to see out of the second gen scarlet, and the base line model? Also, what could be usefuld for the vast majority of non motion picture professionals and small independents?

 

[Dan Hudgins]

Epic sensor cost+olpf+dram

Epic sensor cost+olpf+dram

I'm still wondering what the cost of the Epic sensor+OLPF+DRAM buffer for 30 second shots would be? The image quality would be equal or better than EPIC and the cost should be 1/4 or less, for people who want the look without the price...

 

[David Rasberry]

The uncompressed Bayer raw data rate for Scarlet 3k 16bit per pixel at only 24fps is almost 240MB/sec( that's megabytes not bits). At the lowest redcode compression, 3:1, which is mathematically lossless so no frame data is compromised at all you are still looking at 80MB/sec sustained write to the capacity of the drive. Black Magic makes a cheap uncompressed 1080p 10 bit video recorder, but it only accepts encoded video and at the moment only works reliably with one or two particular models of Intel SSD's that can sustain high continuous write speeds for the entire capacity of the drive. Many SSD's slow down considerably as they fill up.

All I saw on Ambarella's site was various kinds of H.264 codec chips. Not really an acceptable codec for high quality image acquisition as it does not hold up well for post production or give you much room for image manipulation without breaking down. Fine for release formats or video conferencing. This is what the current gen DSLR's already record and the results are mediocre.

As to Scarlet 2, I would like larger sensors, maybe 4/3 but at least S16mm, at least 8 mega pixel QHD resolution, with improved sensitivity and dynamic range.

 

[Michael Olsen]

The real question for Scarlet 2 buyer will be, I think:

New Scarlet 2 or used Epic. Humm...

Then again, given the upgradability and the already looming concept of DRAGON, perhaps there won't be many used EPICs on the open market.

 

[conrad gaunt]

Wavelet compression isn't necessarily computer intensive (analysis takes place in the spatial domain, without the computationally intensive transformations required when doing analysis in the frequency domain), the math can be quite simple. Bandwidth is a bigger concern when building the signal/recording chain, and of course recording media speed. Jpeg2000, redcode etc, aren't the only wavelet codecs.

Dan, stop shooting 24fps.. 25fps is a better option for internationally released cinema, so better to at least support that fps option when building an imaginary system IMHO

How about aiming for more dynamic range and less resolution. 2k with quad pixels?

 

[Dan Hudgins]

2K Bayer vs. 2K RGB

2K Bayer vs. 2K RGB

How about aiming for more dynamic range and less resolution. 2k with quad pixels?

I've been working with various True RAW Bayer data for the last three years, and have to say that for theatrical projection from Bayer data it's best to have at least 2x 1280x720 since 1280x720 is about what you see projected off a 35mm print. So ideally the sensor should have at least 2560x1440 pixels if it is a Bayer camera, that is the most important issue, more so than dynamic range since good results can be had from sensors with 12bits though the use of area filters and extended range processing in the de-Bayer.

We transferred some 1280x720 to 35mm and projected it, and its close to a 4th generation 35mm print, just a little loss of resolution, the main issue is aliasing and Bayer artifacts (as well as compression artifacts and noise).

ARRI and Scarlet use the 1.5:1 reduction trick to null some of the Bayer artifacts, so shooting 1.5x at 2880x1620 or 2x at 4096x2160 should be the target for the design of new cameras, or more current shoot 6K for 4K or 8K for 4k if you are using a Bayer camera and a RGB DCP projector.

RED has does a great job with its de-Bayer for 1:1 use, but most of their 4K ends up as 2K or HD so its reduced about 2x most of the time anyway and so that nulls much of the residual Bayer artifacts. Future versions of Epic should use a 6K sensor for 4K end use, as 5K does not null when reduced to 4K.

In case what I was talking about was not understood, in a BUFFERED camera the images are recorded for ONE SHOT into the camera's buffer, any frame rate can be supported it just increases the cost of the buffer to support higher frame rates and shots longer than the minimum needed of 30 seconds, and if you are making a minimum cost option, then you go for the biggest "bank for the buck" and use a good RED sensor in a buffered camera head.

The bandwidth going to the HDD for storage is nothing to consider other than its turn-around time for the dump cycle after the shot is finished or the buffer is full and needs to be dumped or swapped. If the camera has two buffers, then one can be used to shoot another shot while the other is being dumped so you have less or no wait before the next take.

The advantage is that you can use low cost 2TB HDD for storage of the frames and the bandwidth does not matter any disk even slow ones will work just fine and have no impact on the frame rates you can shoot since ALL shooting goes to the high speed DRAM or RAID FLASH buffer. In the instance of using 24 or 48 CF cards with a multi phase clock it lets the data be distributed to the buffer at much lower speeds than the sensor operates since you divide the sensor's data rate by 24 or 48.

The individual pixel dynamic range of the sensor it self does not matter since area processing is standard for de-Bayer now and can reduce the noise to about the same for sensors of equal area even if one is large pixel and the other is small pixel, except the small pixel sensor is easier to de-Bayer with fewer artifacts. Much of what makes one brand of camera look better than another is not the sensor but the way the sensor's data is processed as even the best sensors have some noise that prevents their output from being used without processing at least for the matrix-ed chroma to reduce noise.

Using a 1920x1080 pixel sensor is not a good idea for 1920x1080 results as its almost impossible to hide all of the de-Bayer artifacts. If you use a Quad-HD sensor and produce in-camera de-Bayer 2x then you can record RGB, but that gives 3x the bandwidth of 1920x1080 Bayer data, but 3/4 the data of the Quad-HD 3840x2160 Bayer. So in that case, it might be better to do in-camera de-Bayer and save the data RGB if you only need 1920x1080 results from a 3840x2160 Bayer sensor. Pixel Binning is another issue as that gives you a Bayer pattern, rather than RGB result, so makes the Bayer artifacts larger rather than nulled.

When you use many SSD to store minutes of uncompressed HD data you do have a high cost storage media, but if you use a smaller buffer that just holds one or two shots it's cost is lower, and you immediately dump the shot or shots to low cost HDD media so you don't need to copy the SSD and so that step and time are eliminated. The speed of the HDD does not matter much so the lowest cost and slowest HDD will work just as well as faster higher cost ones other than for the dump time which the end user can decide upon for his budget.

One of the complexities of FLASH memory is the need to map the bad addresses and that adds to the cost and complexity of the FLASH memory and its maximum write speed. DRAM can be 100% good addresses, and so does not need as complex methods to correct for the work-arounds of the bad memory locations, also DRAM can write faster and at a sustained rate whereas FLASH does not write at an even rate and is more complicated and slower to re-write.

Dynamic range is a product of the image as a whole and not just of each single pixel when viewed at normal distances, so even sensors with higher noise levels can give 12 stops or more of tonal range for patches larger than single pixels.

 

[Wayne Morellini]

..one or two particular models of Intel SSD's that can sustain high continuous write speeds for the entire capacity of the drive. Many SSD's slow down considerably as they fill up.

All I saw on Ambarella's site was various kinds of H.264 codec chips. Not really an acceptable codec for high quality image acquisition as it does not hold up well for post production or give you much room for image manipulation without breaking down...

As to Scarlet 2, I would like larger sensors, maybe 4/3 but at least S16mm, at least 8 mega pixel QHD resolution, with improved sensitivity and dynamic range.

Hmm, yes. Interesting that SDD's slow down, I am unfamiliar with all the new fangled technology, I would have not expected this from silicon, this is what we get from hard disks. Does pre-assigning sequential recording space improve this?

The ambarella is programmed for h264, but the model I quote in particular canbe programmed for a different codec. As Conrad said, particularly compared to h264, wavelet canbe not as intensive as it seems. Jpeg2000 was a big hog, and cineform was much lower in processing requirements. Gpu programming also brings up a lot of power, and the Ione has a signifcant Mobile Gpu and it's own processing arrays, not what I would prefer as a Gpu, but reasonable for what you can get mobile wise.

Wavelet compression isn't necessarily computer intensive (analysis takes place in the spatial domain, without the computationally intensive transformations required when doing analysis in the frequency domain), the math can be quite simple. Bandwidth is a bigger concern when building the signal/recording chain, and of course recording media speed. Jpeg2000, redcode etc, aren't the only wavelet codecs.

Dan, stop shooting 24fps.. 25fps is a better option for internationally released cinema, so better to at least support that fps option when building an imaginary system IMHO

How about aiming for more dynamic range and less resolution. 2k with quad pixels?

Yes binned. I would like to know what the fillfactor and shape of fillfactor of the scarlet pixels.