Will there ever be RAW codec for Canon HDSRLs
- Thread starter tackyattack
- Start date
1. The data rates required for raw motion are much too high to record to any currently available off the shelf solid state media. We are at least 3 Moore's law cycles away from the technology to do this at DSLR price range and form factor. 3 to 5 years minimum.
2. It won't likely ever happen in consumer price range cameras because the average consumer doesn't want to store and deal with terabytes of data to make home movies or deal with the processing of that data into a usable format.
The sensor data recorded is only directly affected by light. aperture, and exposure time defined by frame rate and shutter speed. All other settings that affect how the processed deBayered image looks are meta data applied non-destructively either in camera or in post and subject to change at will. Clip meta data is recorded to a separate file from the raw image data file and has the same file name but a different extension, .RMD instead of .R3D.
2. It won't likely ever happen in consumer price range cameras because the average consumer doesn't want to store and deal with terabytes of data to make home movies or deal with the processing of that data into a usable format.
Redcode raw is full 12 bit or 16 bit per pixel Bayer raw data recorded frame for frame to a file unmodified except for a very sophisticated data compression scheme that allows full recovery of the raw data for post processing. At the lowest compression rate it is mathematically lossless, meaning no data for any pixel is discarded. At the highest compression rate it loses a very slight amount of high frequency detail, but this is rarely visible at full resolution and not visible at all at HD video resolution. It is as raw as any frame that comes from a Canon or Nikon still camera.
The sensor data recorded is only directly affected by light. aperture, and exposure time defined by frame rate and shutter speed. All other settings that affect how the processed deBayered image looks are meta data applied non-destructively either in camera or in post and subject to change at will. Clip meta data is recorded to a separate file from the raw image data file and has the same file name but a different extension, .RMD instead of .R3D.
Last edited:
Samuel H
Major Contributor
* SSDs are capable of sustaining the 208MB/s that came from my calculations on the previous page (edit: after redoing the math, just by a hair, but still there), even without inter-frame compression; you just have to forget about 5K
http://www.anandtech.com/show/4202/the-intel-ssd-510-review/5
* an update on my numbers:
forget the "nice binning" step, I don't need to store the whole bayer pattern stuff: I want super-clear 1080p, with the best possible color information, so instead of that "nice binning" I'll just debayer, and from that poing on keep just an RGB 1920x1080 14bit image (with "sensor color space", meaning each the units on each RGB channel are exactly the same as in the original bayer pattern)
http://www.anandtech.com/show/4202/the-intel-ssd-510-review/5
* an update on my numbers:
forget the "nice binning" step, I don't need to store the whole bayer pattern stuff: I want super-clear 1080p, with the best possible color information, so instead of that "nice binning" I'll just debayer, and from that poing on keep just an RGB 1920x1080 14bit image (with "sensor color space", meaning each the units on each RGB channel are exactly the same as in the original bayer pattern)
Last edited:
Danielvilliers
Veteran
I don't understand how people can say never with technology. Just look at what you where using five or ten years ago to see how everything has changed. Tens years ago we were all using film. Nikon came with the revolutionary D1 at 2.7 mega pixel and + $ 5000 price range. Tell someone in those days that you would get a 18 megapixel camera for less than $ 800. I am not even talking about low light etc. What was the processor and storage in those days, I prefer not to think about it. If you told someone that he would have to store a 25 megabyte still raw photo on his computer, he night tell you that it will never happen.
For me we will get at first a 1.5 dslr revolution. It will come in the form of clean uncompressed hdmi output dslr that can be recorded with an external recorder. Many of the flaws of the dslr are already taken care of by the Panasonic gh2 (overheating, no time limit, resolution, moire and aliasing greatly reduced, autofocus). It just need a clean hdmi to become the true 1.5 revolution. Canon is not stupid it will not undermine its whole dslr line by putting it in something like the 60D. We will surely see it in the Canon 5d mark 3, then it will tickle down in the successor of the 7d etc. They are business after all which want to make profit and encourage you to upgrade.
Then we will see a true second dslr revolution or accident. Someone will implement a 24 fps raw output mode. I say mode because it might just be a set of still image without a proper video codec etc. The gh2 has a 40 fps jpeg 4 mega pixel 1 second burst mode for example on the gh2. It might not be full raster, it might use some compression etc. but it will come. It might be handy for sport and journalist photographers that will be able to make continuous filming then extract still frames from them. A 4/6 megapixel version will be enough for many publication.
When I see the very strong rumors of an April launch of Nikon mirroless professional camera it is very exiting. Many have been questioning Nikon for not engaging the video market because it had no professional video camera market to protect. The answer is perhaps coming. They have patented a lot of technology relating a lot to video. From pelicle mirror like in the sony nex line to autofocus for video. It has taken a lot by surprise that they are targeting the high end rather than the low end with their mirror-less camera.
I say that we will find a version of raw video/still output from dslr in the next 3 to 5 years. It first implementation might not be the most practicle at first, but further implementation will make them viable tools for filmaking not long after 2015. By that time, every middle range PC will have from 8 to 12 processor core with 2 terabyte HD as standard. What is good about the dslr market is that they are coming from a fresh start, so they will evolve much more rapidly than the video manufacturers that have already segmented there market. The same dslr revolution 2 will force the video world to implement raw much faster. Do anyone think that we would have the Panasonic Af-100 or Sony coming S35 NXCAM at the $ 5000 price range without the dslr video revolution.
For me we will get at first a 1.5 dslr revolution. It will come in the form of clean uncompressed hdmi output dslr that can be recorded with an external recorder. Many of the flaws of the dslr are already taken care of by the Panasonic gh2 (overheating, no time limit, resolution, moire and aliasing greatly reduced, autofocus). It just need a clean hdmi to become the true 1.5 revolution. Canon is not stupid it will not undermine its whole dslr line by putting it in something like the 60D. We will surely see it in the Canon 5d mark 3, then it will tickle down in the successor of the 7d etc. They are business after all which want to make profit and encourage you to upgrade.
Then we will see a true second dslr revolution or accident. Someone will implement a 24 fps raw output mode. I say mode because it might just be a set of still image without a proper video codec etc. The gh2 has a 40 fps jpeg 4 mega pixel 1 second burst mode for example on the gh2. It might not be full raster, it might use some compression etc. but it will come. It might be handy for sport and journalist photographers that will be able to make continuous filming then extract still frames from them. A 4/6 megapixel version will be enough for many publication.
When I see the very strong rumors of an April launch of Nikon mirroless professional camera it is very exiting. Many have been questioning Nikon for not engaging the video market because it had no professional video camera market to protect. The answer is perhaps coming. They have patented a lot of technology relating a lot to video. From pelicle mirror like in the sony nex line to autofocus for video. It has taken a lot by surprise that they are targeting the high end rather than the low end with their mirror-less camera.
I say that we will find a version of raw video/still output from dslr in the next 3 to 5 years. It first implementation might not be the most practicle at first, but further implementation will make them viable tools for filmaking not long after 2015. By that time, every middle range PC will have from 8 to 12 processor core with 2 terabyte HD as standard. What is good about the dslr market is that they are coming from a fresh start, so they will evolve much more rapidly than the video manufacturers that have already segmented there market. The same dslr revolution 2 will force the video world to implement raw much faster. Do anyone think that we would have the Panasonic Af-100 or Sony coming S35 NXCAM at the $ 5000 price range without the dslr video revolution.
Last edited:
You will never run Crysis on a Pentium II.
The problem is not that it can't be delivered. The problem is that it won't be delivered at a price that you can afford at this moment in time. Of course, on a long enough timeline, all prices drop steeply, and, hey, in 50 years maybe people will be able to afford a 4k RAW video camera for less than $1000, but for the workable future that is not going to be the case.
haha.. I chuckled at that...
Darren Levine
Veteran
LMAO, but what if you hacked the core of the PII and supercharged it with NOS and racing stickers, then, maybe then?
it's the same stupid debate of people wanting everything but to pay nothing. anyone expecting a 5d3 wth XLR and raw is just plain stupid in my eyes. sorry, have to be stern about this and inject some realism. the camera you want isn't a photo camera and it costs 20, 50, 150, 250 THOUSAND dollars. or just wait another 10 years.
i'm just tired of people expecting it thats all, it gets really annoying, especially if i haven't had my tea yet.
it's the same stupid debate of people wanting everything but to pay nothing. anyone expecting a 5d3 wth XLR and raw is just plain stupid in my eyes. sorry, have to be stern about this and inject some realism. the camera you want isn't a photo camera and it costs 20, 50, 150, 250 THOUSAND dollars. or just wait another 10 years.
i'm just tired of people expecting it thats all, it gets really annoying, especially if i haven't had my tea yet.
Samuel H
Major Contributor
no: RAW is "relatively unprocessed sensor data", 4:4:4 is "no color subsampling"
plus, another update on my numbers (please, imdjay, skip this, it will annoy you
)IF the buffer is just being used for delaying the final write-to-the-card process (that's a big if, but see * below), then my 550D can move around data at a whooping maximum speed of 680 Mbps (25MB RAW files running at 3.4 fps), which is (edit) LOWER than my 1250 Mbps for a stream of 24fps debayered 1920x1080 RGB-14bit images, without inter-frame compression
so moving the data around would (edit: take the "not" away) be the bottleneck
apart from this, what we need is:
* a faster sensor, capable of taking pictures at 24Hz
* really fast debayer, to convert the bayer pattern from the sensor into a sharp 1920x1080 RGB image with 14-bit color, using the sensor's RGB color space (i.e. each the units on each channel are the same as in the original bayer pattern)
* fast and big storage media (read: SSD), to keep the buffer from getting filled, and allowing some decent recording times (20 minutes on a 256GB drive)
again: totally feasible, you just have to work with that goal from the start; I still bet on hardware currently at the start of its design process, and so 2013
* dpreview says the 550D can take 6 RAW stills, at 3.4fps, before the buffer fills, and then you have to wait 7 seconds for the write operation to be completed; that's totally coherent with the 9 seconds that the 8 GB Sandisk Extreme III card they're using would take to write those 150MB, at the 16MB/s speed that this test achieved
http://www.legitreviews.com/article/1172/2/
so the SD card is the biggest bottleneck, and, even though that doesn't mean it is the only one, it means that IF is still big but just not insurmountably big
Last edited:
Danielvilliers
Veteran
Many here said never, English is not my first language but I thought that never meant (not at any time, dictionary). The OP did not specify if it will happen today and I myself did not say that it will happen now. But in the next 3 to 5 years I see a version of raw output coming out of a Dslr. If you see some technological site you will see that many tech are being develop that will increase a lot the speed of CF and SD card. As NormanBates calculation state internally the data throughput are already their in an entry level camera. The 7d is already doing twice what his 550D is doing. I am not encouraging anybody to just wait for the next best thing and not to shoot. Canon just need to put some clean hdmi output and get rid of the moire aliasing as in the gh2 for the 5d mark 3 to be competitive in the video market.
Now tell someone 3 years ago that they would be shooting vista vision size video for about $ 2500... two years ago.
Kevin Marshall
Active member
Gonna make a bit of a semantic point here, but I think it's still important. RAW and Uncompressed are different things. You can have compressed RAW, and you can have Uncompressed RGB/YUV/whatever. Uncompressed simply means that - the image hasn't been compressed. RAW means the sensor values have not been interpolated to a color-encoded image prior to compression/recording.
They still aren't.
Danielvilliers
Veteran
The full frame Canon 5d mark 2 sensor is about the same size than 65mm vista vison. In terms of resolution and other film characteristic it is not. But I guess you understand my point.
morgan_moore
Major Contributor
havnt read the whole thread but..
2002 nikon D100 - time to write a 6mp raw file - 40seconds - card size 64-128mb
2005/6 (?) nikon D3 time to write a 13/mp raw file 1/8 of a second card size 8gb
2013?
i cant see it not happening.......... sometime
S
2002 nikon D100 - time to write a 6mp raw file - 40seconds - card size 64-128mb
2005/6 (?) nikon D3 time to write a 13/mp raw file 1/8 of a second card size 8gb
2013?
i cant see it not happening.......... sometime
S
I think you'll find that canons brand name is actually just a arm CPU brand name for marketing. It's the same chip. Just with a marketing brand name on it.
Perry Wilson
Well-known member
Actually word has it the replacement for the 5D MK2 will have raw video capabilities... not my words just what have read...
robotsound
Well-known member
If Wikipedia is to be believed, the DIGIC 4 chip is an ASIC, which contains an ARM CPU, as well as other application specific instructions and support, implemented in hardware.
http://en.wikipedia.org/wiki/DIGIC
http://en.wikipedia.org/wiki/ASIC
I would hazard a guess here that if Canon cameras are anything like Cisco routers (which also use ASIC chips in some, but not all, of their products), they are probably doing as much of the heavy lifting as possible in dedicated custom hardware. All things being equal, a hardware implementation is always much faster than a pure software implementation.
It would seem that DIGIC 4 is probably not just an ARM CPU running code, although an ARM CPU is a component of the system.
http://en.wikipedia.org/wiki/DIGIC
http://en.wikipedia.org/wiki/ASIC
I would hazard a guess here that if Canon cameras are anything like Cisco routers (which also use ASIC chips in some, but not all, of their products), they are probably doing as much of the heavy lifting as possible in dedicated custom hardware. All things being equal, a hardware implementation is always much faster than a pure software implementation.
It would seem that DIGIC 4 is probably not just an ARM CPU running code, although an ARM CPU is a component of the system.
Thanks Gary, I agree with you and my point I was trying to make when I said never. I knew when I said that someone else would not understand my point. If they get to that goal I doubt they'll want to do all that work just to throw it on a DSLR. If they did all that it would just be a high end movie camera and not a DSLR, like I said earlier like a RED.
Samuel H
Major Contributor
@ Rexyinc & robotsound:
since digic II, digic processors have both an ARM part that is general-purpose and fully programmable, and an ASIC part that is much more task-specific (which means the whole chip can be categorized as an ASIC)
everything magic lantern does is run on the ARM part (drawing stuff on the LCD, controlling the camera settings, activating or deactivating various parts of the camera, etc); the bit that the ASIC does is processing the information from the sensor and getting out of it your pretty images; there's no knowledge outside canon on how this ASIC part is programmed, that's why ML can't change things like image resolution, line skipping, etc
since digic II, digic processors have both an ARM part that is general-purpose and fully programmable, and an ASIC part that is much more task-specific (which means the whole chip can be categorized as an ASIC)
everything magic lantern does is run on the ARM part (drawing stuff on the LCD, controlling the camera settings, activating or deactivating various parts of the camera, etc); the bit that the ASIC does is processing the information from the sensor and getting out of it your pretty images; there's no knowledge outside canon on how this ASIC part is programmed, that's why ML can't change things like image resolution, line skipping, etc
This is the primary reason. Don't expect modestly priced consumer still cameras to morph into movie cameras. It is not the primary purpose or market.