The holy grail (Canon Raw Rumor)

[Barry_Green]

I just think that maybe Red underestimated how much people are willing to "give in" when it comes to the price / performance / image aesthetics ratio.

It's not just Red, and it's not just imaging. It's the whole world, and it's in everything. It's the "good enough" revolution, in action.

All we ever really wanted was an inexpensive HD camera with a 35mm sensor. Isn't that why everyone was building and buying 35mm adapters for their DV/HDV/P2 cameras all these years? The new DSLR's aren't perfect, but for the price... they offer what we've been after.

Yep, especially when the price gets cut to below a grand. That's exactly why I said that even though the DSLR, as a professional video camcorder, pretty much sucks -- it's plenty good enough that all the folks waiting patiently for the Scarlet could now get off the fence and start shooting something. 'cause, for most of us, the DSLR video is plenty "good enough" (see paragraph 1).

For those of us for whom it's not good enough... that's what the S35 Scarlet is for. For the few of us who will pay the ten grand for "perfect" imagery -- it's going to be a viable product, and better than an SLR in every way.

Except one -- price. It'll cost about 15x to 20x as much as a t2i. And for whole swaths of the population, that t2i or 7D or GH1 or >whatever< is gonna be "good enough."

 

[David Del Real]

It's not just Red, and it's not just imaging. It's the whole world, and it's in everything. It's the "good enough" revolution, in action.


Yep, especially when the price gets cut to below a grand. That's exactly why I said that even though the DSLR, as a professional video camcorder, pretty much sucks -- it's plenty good enough that all the folks waiting patiently for the Scarlet could now get off the fence and start shooting something. 'cause, for most of us, the DSLR video is plenty "good enough" (see paragraph 1).

For those of us for whom it's not good enough... that's what the S35 Scarlet is for. For the few of us who will pay the ten grand for "perfect" imagery -- it's going to be a viable product, and better than an SLR in every way.

Except one -- price. It'll cost about 15x to 20x as much as a t2i. And for whole swaths of the population, that t2i or 7D or GH1 or >whatever< is gonna be "good enough."

Nailed it!
QFT.

I have three words to say about it all: budget, budget, budget. What matters to me most is getting the best shot with the best equipment I can afford - NOW. Was saving for a 7D but now have my eyes on a pair of T2i's thank you very much.

 

[Chris Santucci]

Was saving for a 7D but now have my eyes on a pair of T2i's thank you very much.

Now... that's crazy talk. :cheesy:

.

 

[Ryan Farnes]

The most interesting scenario I sort of see unfolding is a DSLR that shoots RAW 1080p video, released within months of the Scarlet, at a price significantly less than an S35 Scarlet. The 7D is a mirage right now compared to the RED ONE.

However, a RAW video DSLR a year or less from now (who knows) could be a serious contender against all the RED cameras, especially if it masters to 1080p on par with most or all of them. We've seen 7D/5D/GH1 footage cut against RED at 720p on the internet (cough), but no, seriously, that stuff can cut against it when scaled down. No reason to doubt that the trend holds over the next couple of years to 1080p, particularly with no where to go but up with the DSLRs.

Maybe that is in part why Jim is pushing 4K mastering? Talking up RED RAY? Not so much the whole DSLR thing, but the whole 1080p thing. His cameras have a distinct advantage when it comes to future proofing at high resolution. Problem is, in a world still waiting for Blu-ray saturation, 2K digital cinema saturation, 1080p satellite and cable saturation, 1080p TVs in most homes, and 1080p internet distribution....

....1080p may be king for a few years to come. It will all come down to what you and your clients want.

 

[ChipG]

I'll bet in two years we have a RAW T2i or T4i or whatever they call it then for $999. One thing about Canon is they have the means to fast track R&D if they choose at a hyper speed compared to RED.

I thought RED was done with 7D's on their site? This makes me want to see the test redone using a GH1 now LOL

 

[Calomile]

I think (and hope, oh god I hope...) that it's inevitable that DSLRs will eventually shoot raw images (video and still). At the very least, Trammel Hudson of Magic Lantern fame has been making steps into trying to dump the raw sensor data to the camera's compact flash, but as far as I'm aware he hasn't made anything viable for cinematographers (his words, in a recent interview.) - this might have pushed Canon into thinking "Damn, we'd better push something out before he discovers something we don't want him to..."

I'm wondering, since obviously raw would use the entire sensor, would this mean there would be less aliasing and moire, since the camera will no longer be line skipping?

 

[Ian-T]

I'm wondering, since obviously raw would use the entire sensor, would this mean there would be less aliasing and moire, since the camera will no longer be line skipping?

I'd think there would be none of the above...just like when you take a RAW photo.

 

[jamesmallonuk]

Wow so were going to get RAW pixel accurate glorious 2-4K Jello, Nice.

 

[Michael Olsen]

I'm wondering, since obviously raw would use the entire sensor, would this mean there would be less aliasing and moire, since the camera will no longer be line skipping?

This is a good question because it leads to me think of considerations.

If we all woke up tomorrow to find that our 7Ds could record RAW, our camera would have essentially become a 5K RAW camera with an ~s35 sized sensor. Wahoo!

However, every frame of video would take up between 17-22 Megabytes (the same as 1 RAW still). That's per frame.

So one second of footage at 24 frames per second would take up about 468MB (19.5MB average per frame*24 frames in one second=468MB per second).

That means that in order to record one minute of footage, you are looking at approximately 28GB of information.
(468MB per second*60 seconds per minute = 28,080MB per minute or ~27.42 GB)

This raises two very real problems.
1) What storage medium are we going to record all of this to? HDD is the only thing which can currently support that disgusting amount of information. Which leads to other problems (dropped frames, power, etc).

2) Recording speed. What do we have that records at something close to 468MB/s? 667x CF cards top out at 100MB/s...and that probably isn't sustained. In order to actually record something spitting out information this quickly, we'd need a pretty nice (and sizeable) RAID array.

So what are the solutions to this mess?

Keeping it RAW:
1) Window the sensor. Only shoot out of a 2K window of the 7Ds sensor so that filesizes and recording requirements drop. But this virtually entirely negates the s35 size of the chip...it becomes about a 1/2" chip in terms of crop factor and FOV.

2) Make a new sensor that is ~2K and still S35. This means having either really large photosites, or having significant amounts of space between photosites. My understanding is that either of these isn't really a great option due to both high cost and potential performance problems (screen door effect, noise problems).

3) Use 3 CCDs and a beamsplitter. This is really, really expensive.

4) Lineskip to get to 2K. But this is where we are now. And no-one really wants RAW if it is going to be aliased and speckled with moire.

Keeping it RAW - sort of:
1) Compress it. Use a form of compressed RAW like sRAW or mRAW. This can help mitigate filesizes, but does decrease the amount of information recorded. And (to the best of my knowledge) these forms of compression are made for use with still images...they may not work so well with motion. Using an improper form of compression can lead to similar effects that we are already seeing - moire, color shifts, macroblocking, and aliasing.

Unfortunately, using a good compression scheme for ~3-5K footage requires some pretty significant processing power in the camera. In fact, the ASICS for the new REDs do this job. And they use, by far, the most complex ASICs ever created*.

Forget RAW, just give us great 1080p:
1) Don't skip lines or pixel bin in the recording, just use in-camera debayering to deliver out RAW. Unfortunately, this takes a LOT of processing power that DSLRs just don't have. Really, even most video cameras don't have the guns to do this.

*According to Jim, who is speaking on behalf of the ASIC engineers.

 

[Barry_Green]

I'm wondering, since obviously raw would use the entire sensor, would this mean there would be less aliasing and moire, since the camera will no longer be line skipping?

Your fundamental premise is incorrect. Raw can't possibly use the entire sensor, because the sensor cannot be read out fast enough.

The aliasing and moire don't come from the codec! They come from the way the sensor is read. These cameras have to skip 85% of the pixels on their sensor to speed up the readout rate to video-compatible speeds.

If you recorded a line-skipped, pixel-binned video system in raw, you'd have a more-accurate, clearly-visible line-skipped, pixel-binned shot. It won't reduce aliasing or moire at all (because those are inherent in the sensor), it'd probably just make them more clear and visible.

 

[Michael Olsen]

Your fundamental premise is incorrect. Raw can't possibly use the entire sensor, because the sensor cannot be read out fast enough.

Didn't even think about that limitation, Barry. 7D sensor would have to read out 4 times faster than it does now to get us to 30p...

 

[daveswan]

I think the assumption is that the sensor and computing is fast enough to read the full raster without line skipping.
At least that's my reading, could be wrong of course :smile:
Dave

 

[shaocaholica]

Keeping it RAW - sort of:
1) Compress it. Use a form of compressed RAW like sRAW or mRAW. This can help mitigate filesizes, but does decrease the amount of information recorded.

This is interesting. Canon's sRAW on the 7D is surprisingly good. I don't know what algorithm they are using but the end result is significantly smaller yet very sharp and for some reason in my tests, has a better noise profile than the original scaled down. Even if they just stored individual sRAW, it would be incredible. Maybe add a little more compression and keep it intraframe. The bigger step would be to come up with a motion compression algorithm but I think sRAW is still the way to go since its much more size managable and still larger than 1080p. If anything, we'll see sRAW type RAW video first, at least from Canon, before we see full rez RAW video. Which is fine by me since I don't need to go beyond 1080p for my projects.

Edit:

Here's some sRAW math.
~12mb/frame (max) 2592x1728
@24fps = 288mb/s
Thats a lot of bandwidth but you'd be getting 2.5K RAW. Lets assume that we want a target of 100mb/s which is realistic for short term future CF cards. You would have to bump up compression by 3x on those sRAWs.

 

[Michael Olsen]

Just another consideration...

Would RAW and/or sRAW/mRAW require debayering prior to editing? I know REDRAW goes through REDCINE-X for a debayer and encoding to an intermediary...

Wasn't there some sort of open source motion RAW format relatively recently announced or supported by Adobe? CineRAW or something like that?

 

[roxics]

What if they don't care about aliasing and moire and just output more color information?

Or what if they increase the resolution twice that of 1080p and just down sample in camera to 1080p? Would that decrease the effect?

Either way I don't think Canons idea of raw would really be raw. If anything it would probably be their new 4:2:2 format they recently announced.

Even if that's the case, that would wonderful!

 

[shaocaholica]

Just another consideration...

Would RAW and/or sRAW/mRAW require debayering prior to editing? I know REDRAW goes through REDCINE-X for a debayer and encoding to an intermediary...

Wasn't there some sort of open source motion RAW format relatively recently announced or supported by Adobe? CineRAW or something like that?

Depends on how the camera maker wants to implement it. If you resize a RAW, its technically not RAW anymore but you could still keep in a bayer pattern.

Here is a snippet on Canon's implementation from a 3rd party:

http://www.barrypearson.co.uk/articles/dng/specification.htm#sRAW

" The SRAW image is in a hybrid format; it is not truly raw any more. It is raw only in the sense, that the colors are in the camera's color space and WB has not been applied yet, but otherwise it has nothing common with the original raw format. It is demosaiced, i.e. the pixels have three color components. However, not each pixel has three separate color component: adjacent pixel pairs are sharing two of the three color components."

Maybe the description "partially demosaiced" might be appropriate for the layman.

 

[Michael Olsen]

Depends on how the camera maker wants to implement it...

Interesting...I never really looked much into exactly what sRAW does. So even with sRAW there would still be a partial debayer necessary in order to create a real RGB file for editing.



I just did a little bit more math looking at sRAW. It is supposed to provide a file of 33% to 50% smaller than RAW. I'll take 40% as a nice middle ground that slightly favors smaller files.

(19.5MB RAW frame * .4 sRAW conversion = 7.8MB sRAW frame)
(7.8MB sRAW frame * 24 frames per second = 187.2MB per second)
(187.2MB per second * 60 seconds = 11232MB per minute = 10.98GB)

So we are still looking at a lot of data moving very very quickly. More than any easily portable solutions I know of can handle. I think Canon would have to look at creating their own RAW compression.

 

[shaocaholica]

I found a sRAW resolution chart from the 7D at www.imaging-resource.com. Processed it through LR Beta 3 default settings.

To clarify, this is a crop from a 7D sRAW still (2592x1728) not from video

You still have some moire and aliasing but its obviously way better than lineskipping. Since the frame size is larger than 1080p, you could actually blur then resize to remove some the artifacts without loosing sharpness (I think).

 

[Michael Olsen]

What if they don't care about aliasing and moire and just output more color information?

Then I think we'd basically get a 4:4:4 camera that outputs an H.264 encoded file that actually resolves around 720p, but uses "false detail". It would still alias and moire just as it does now.

Or what if they increase the resolution twice that of 1080p and just down sample in camera to 1080p? Would that decrease the effect?

Shoot 4K and finish 2K is a pretty decent idea - really, it's how a lot of people work on the RED. And the extra $16000 price difference isn't just for profit. Making a 4K sensor that can read out 24 or 30fps is expensive, as is putting in the hardware and software that can take a 4K image and reliably downscale to 1080p 30 times a second, every second, for a long time. And then you've got either A) even more expensive powerful hardware to do a good RGB conversion or B) expensive and likely sizable storage to record the RAW files. If you think heat is a problem with the 7D now...yeesh! This kind of camera would pretty much have to have a fan, some really righteous external heat sinks, or be water cooled.

Either way I don't think Canons idea of raw would really be raw. If anything it would probably be their new 4:2:2 format they recently announced.

I'd be very curious to see what they've come up with to solve the storage problems I mentioned above. Fibre channel SSD array, perhaps? Wonder how much that would cost...

I hope that my commentary isn't coming across as negative, but rather as realistic. There are some very significant hurdles in terms of RAW recording that have to be overcome in order to create something that is useful without being extremely cumbersome. I really do hope Canon has some ace up their sleeve to provide a great RAW camera. But I really don't see how they could do it without a number of rather expensive innovations. And I'd think that, at this point, that they would have at least announced or hinted at it officially. If for nothing else, than to discourage people buying into RED and considering them "innovators". After all, as soon as 2 people are doing something it stops being unique and starts becoming "industry standard".

 

[shaocaholica]

Canon could make a vertical grip type add-on that takes standard 2.5" SSDs which are something like 150-200mb/s sustained write. If you don't want it, you don't pay for it.