Canon APS-C sized Video Camera

[combatentropy]

So, are you saying it is possible?

.... To readout 20 Million pixels from a CMOS chip, do a full resolution debayer process, using every single pixel on the CMOS chip, without pixel binning, then scale the resulting images, 30 frames a second, to a standard broadcast resolution, like 1080P? And, simululateously output that signal to a viewfinder on the camera, the HD-SDI port for display on a monitor, and to an MPEG-4 encoder and record the video on a CF card? Simultaneously?

That is what it would take to make a functional video camera, right? Being that I am more of a cameraman and not an engineer, clearly it is not for me to say what is and isn't possible with today's technology. I can only point out that, that the most capable video camera on the market today (RED ONE) does not even do what you're talking about... For some reason. And, my understanding is that reason is because it's not possible.

Well, first of all, its 12, not 20, million pixels. The Nikon D90, for example, reads, debayers, downscales, compresses, and writes at 24 FPS. And I doubt that this thousand-dollar camera is at the cusp of what is possible.

The RED One does not use ASICs. It uses, I think, Mac G4s. RED's big news last December was that it was going to put ASICs in Scarlet, Epic, et al.

 

[Lee Wilson]

The Nikon D90, for example, reads, debayers, downscales, compresses, and writes at 24 FPS.

I don't think the D90 downscales its sensor image.

 

[thisiswells]

Well, first of all, its 12, not 20, million pixels. The Nikon D90, for example, reads, debayers, downscales, compresses, and writes at 24 FPS. And I doubt that this thousand-dollar camera is at the cusp of what is possible.

Ok, I understand now. You're speaking from the point of view of someone who does not own the camera in question. I do. I own a 5D II. And, I have done 4 RED ONE shoots in the past 6 months. I am intimately familiar with the image quality differences. DSLR's, in their current form, in particular the D90, do an unacceptable job at " debayer(ing), downscal(ing), compress(ing), and writ(ing) at 24 FPS" because they are skipping entire rows and columns of pixels (ie - NOT a full debayer) in order to create video files.

The entire reason RED ONE does not use ASIC's, if that is what they are called, is because the quality is SO much better, using a full res debayer, in post production.

Unless I am wrong.

 

[HockeyFan12]

Ok, I understand now. You're speaking from the point of view of someone who does not own the camera in question. I do. I own a 5D II. And, I have done 4 RED ONE shoots in the past 6 months. I am intimately familiar with the image quality differences. DSLR's, in their current form, in particular the D90, do an unacceptable job at " debayer(ing), downscal(ing), compress(ing), and writ(ing) at 24 FPS" because they are skipping entire rows and columns of pixels (ie - NOT a full debayer) in order to create video files.

The entire reason RED ONE does not use ASIC's, if that is what they are called, is because the quality is SO much better, using a full res debayer, in post production.

Unless I am wrong.

It's not too surprising that the 5D II has mediocre video. The hv30 has mediocre stills, after all.

So far as I can tell, the big difference between an ASIC and a general purpose processor isn't speed or capability, but efficiency. The ASIC can do one thing very fast and very efficiently; the general purpose processor can do a lot of things, but with less efficiency. Red's new products all use ASICs; the red one does not.

The chips in the 5D II are designed for photos and only secondarily for shooting video. So the video's not great. In an $8,000 camera designed specifically for shooting video, you'd have to assume the video would be far better as the camera would be built from the ground up explicitly for video. The genesis has done 12-megapixels to 1080p for years, and even the EX1 is capable of combining nearly six million discrete sensors to create 1080p footage, which is then compressed in real time.

Canon won't use row skipping in a video camera. They'd opt for a lower resolution sensor in the first place before doing that. Also, this is just a rumor. We don't even know if this is a real product yet, let alone know the accuracy of these specs.

 

[thisiswells]

So far as I can tell, the big difference between an ASIC and a general purpose processor isn't speed or capability, but efficiency. The ASIC can do one thing very fast and very efficiently; the general purpose processor can do a lot of things, but with less efficiency. Red's new products all use ASICs; the red one does not.

Good to know.

The chips in the 5D II are designed for photos and only secondarily for shooting video. So the video's not great. In an $8,000 camera designed specifically for shooting video, you'd have to assume the video would be far better as the camera would be built from the ground up explicitly for video. The genesis has done 12-megapixels to 1080p for years, and even the EX1 is capable of combining nearly six million discrete sensors to create 1080p footage, which is then compressed in real time.

Based on my limited understanding of how these things work, I agree with you 100%.

Canon won't use row skipping in a video camera. They'd opt for a lower resolution sensor in the first place before doing that.

This is based on a premise I happen to agree with, that an image sensor which uses a full debayer produces images that are alias-free. Thus, starting with the correct number of CMOS pixels to match the output format (2 mega pixels, ie - 1920x1080) would result in sharp, alias free video. Or, alternatively, starting with a high number of pixels, and using an incredibly strong optical anti-alias filter in front of the CMOS to reduce / eliminate the aliasing. Either method *should* be an improvement compared to what we see in DSLR's today. The downside, obviously, is that stills would only be 2 MP, because of the strong anti alias filter or the limited resolution CMOS.

Also, this is just a rumor. We don't even know if this is a real product yet, let alone know the accuracy of these specs.

I have my doubts, but would welcome any new cameras which enable me to keep my dominant / innovator positioning in the market I am located in. Bring it on! I will buy the products you create and use them 150 days a year, to earn a living... Thanks!

 

[Sumfun]

The entire reason RED ONE does not use ASIC's, if that is what they are called, is because the quality is SO much better, using a full res debayer, in post production.

Unless I am wrong.

You're wrong.

The reason the Red One didn't use ASICs is because they would have delayed the release schedule. The Red One uses FPGAs (Field Programmable Gate Arrays) that offer better performance than general purpose processors, but not as much as ASICs. The main advantage of FPGAs is that you can keep changing your code as often as you want, and it's just a matter of updating the flash memory in order to update your system. To make changes with ASICs, you would have to fabricate a whole new chip, which can take months.

But now that the camera code is fairly stable, Red can move to ASICs in the Scarlet and Epic. These cameras will have enough horsepower to do the debayer and scaling in camera, and output full frame 1080p.

Also, please note that after you debayer a 4k sensor, you only get a ~3k image. There are some good explanations on this site and Reduser, but you'll have to search for it.

 

[Spartacus]

Don´t we all like a good thread heating up with speculations?:beer:

 

[SPZ]

Hmm, we're all mods now... April's Fools?

 

[DavidNJ]

I will wait to see a RED Scarlet and their actual cost and price. The distribution, warranty, service, and marketing costs of a $3500 camera are probably $2000k.

The Canon people probably designed Digic 4 in early 2007. It was very insightful for them to figure out how to sub-scan the sensor to get 30 frames/sec, and to increase their recording speed to get H.264 long GOP compression under 17:1. For comparison the EX1/3 is MPEG2 long GOP at 21:1 and AVC Intra 100 is 12.5:1, including more chroma information (4:2:2 vs. 4:2:0) and the potential for more low level detail (10-bit vs 8-bit) but I-frame only. Probably just a little better than Canon.

The only thing that didn't make sense in the report was the 12Mp. If it was 8Mp, 4 photosites/pixel, it would be readily able to process and have full chroma sampling in a 1080p frame. The smaller photo sites in the 12Mp sensor, in addition to processing speed issues, also makes it more expensive to make photo sites of the same quality consistency, introducing noise issues.

As far as heat is concerned, the still camera has neither cooling designed for 30 fps nor a form factor well suited to cooling. It was never a design criteria before, the format exists from film SLRs from before I was kid. RED and SI2k do wavelet compression, which is very processor intensive. If Canon does DCT, as Apple does in ProRes, that shouldn't be any worse then current cameras. As has been pointed out, 3 sensor 1080 full scan, as in the EX1/EX3, HPX 500/300 are already at 6Mp photo sites.

What is more interesting, at least to me, is the RAW option. RED and SI2k do this, compressing the raw sensor output. At 10-bit resolution, an 8Mp sensor only is outputting double what the HPX300 is trying to compress now. The new generation of Compact Flash memory is more than fast enough to capture it with only mild compression.

Net...the 'rumor' seems to be quite a viable description of an upcoming camera, a camera they could announce at NAB in 2009 even if it didn't ship until the fall or later.

On a marketing note, the D90 came out, the 5DM2 quickly followed. Now the GH-1 and 500D and announcements from Samsung and Olympus and presumptions about future Canon and Nikon models. The RED Scarlet created a stir. By the time it sees the light of day in volume you may well see a variety of competing cameras from Canon, Panasonic, Sony and others.

If it records RAW video, will it record 1-bit audio?

 

[SPZ]

STOP THE PRESS (again) RED is changing EVERYTHING! (again)

http://reduser.net/forum/showthread.php?t=28434

 

[thisiswells]

You're wrong.

Hey, I'm OK with that

The reason the Red One didn't use ASICs is because they would have delayed the release schedule. The Red One uses FPGAs (Field Programmable Gate Arrays) that offer better performance than general purpose processors, but not as much as ASICs. The main advantage of FPGAs is that you can keep changing your code as often as you want, and it's just a matter of updating the flash memory in order to update your system. To make changes with ASICs, you would have to fabricate a whole new chip, which can take months.

What I said, was that the reason RED did not offer it from the beginning, was because it was not possible. They still had 2+ years of continual development on their technologies before feeling confident enough to lock them in to an ASIC chip. Even today, it is still not possible because it requires an entirely different camera platform and is many years away because of the development process and the economy. And, when it does eventually come in, it will be priced at 40K+.

What this thread is suggesting, is that the same capability will be implemented into a prosumer video camera for 8K, within the same time frame. Let me put it this way, what are the odds of that occurring? If you were a betting man, what would be your wager?

But now that the camera code is fairly stable, Red can move to ASICs in the Scarlet and Epic. These cameras will have enough horsepower to do the debayer and scaling in camera, and output full frame 1080p.

Which, as we all agree, would be awesome. What I don't understand, is why everyone is convinced that is the way things are moving in the prosumer camera price range??

Also, please note that after you debayer a 4k sensor, you only get a ~3k image. There are some good explanations on this site and Reduser, but you'll have to search for it.

Cool! I will look into that. Sounds funny to me..

 

[Kholi]

STOP THE PRESS (again) RED is changing EVERYTHING! (again)

http://reduser.net/forum/showthread.php?t=28434

Sergio, you were expecting anything more? lol. Come on now.

 

[DavidNJ]

"Which, as we all agree, would be awesome. What I don't understand, is why everyone is convinced that is the way things are moving in prosumer camera price range??"

Canon, Nikon, and Panasonic have announced specific cameras with oversampled 4/3rds to FF35 sensors. Numerous consumer and professional cameras process 8M photosites/frame at upto 60 frames/sec.

SDHC is upto 50Mb/s, Compact Flash is over 250Mb/s. Relatively inexpensive and low cost recorders do DCT compression and output upto 160Mb/s in low power portable devices.

Interchangable lesses came by way of the 35mm adapters, but are there on the DSLR hybrids also. A wide stock of lenses exist, autofocus is problematic. However, I can't imagine using continuous autofocus, the chance of it changing unintendedly is seems far greater than any potential benefit. Does anyone here use autofocus?

As far as demosaicing and down scaling, all the cameras...err...don't do well. They don't have the time or horsepower. For serious still work doesn't everyone shoot in RAW? Ditto the scaling; even the stock scalling tools in Photoshop are limited to bicubic with sharpening or 'smoothing' added. For Lanczos, NEDI, etc you need to go elsewhere. There are plugins to Premier Pro and Photoshop with Lanczos and maybe some others. It is the same algorithm problem.

Both SI2k and RED One are capturing RAW with wavelet compression. The SK2k, because of the sensor isn't oversampled actually can output uncompressed RAW (not sure if they include software to capture it though).

So all the pieces are there...just the funding to put them together. What the RED Scarlet and Canon 5DM2 did is place them front and center in the sub $10k price range. Now it is just time until the manufactuers, under competitve pressure if nothing else, respond.

 

[thisiswells]

I would just like to say, "I love this forum!" In less than 18 hours, I went from making grossly inaccurate statements, to being clarified, and now, I have a better understanding of the issues.

Thanks DVXUser!

 

[joe 1008]

"Which, as we all agree, would be awesome. What I don't understand, is why everyone is convinced that is the way things are moving in prosumer camera price range??"

Canon, Nikon, and Panasonic have announced specific cameras with oversampled 4/3rds to FF35 sensors. Numerous consumer and professional cameras process 8M photosites/frame at upto 60 frames/sec.

SDHC is upto 50Mb/s, Compact Flash is over 250Mb/s. Relatively inexpensive and low cost recorders do DCT compression and output upto 160Mb/s in low power portable devices.

Interchangable lesses came by way of the 35mm adapters, but are there on the DSLR hybrids also. A wide stock of lenses exist, autofocus is problematic. However, I can't imagine using continuous autofocus, the chance of it changing unintendedly is seems far greater than any potential benefit. Does anyone here use autofocus?

As far as demosaicing and down scaling, all the cameras...err...don't do well. They don't have the time or horsepower. For serious still work doesn't everyone shoot in RAW? Ditto the scaling; even the stock scalling tools in Photoshop are limited to bicubic with sharpening or 'smoothing' added. For Lanczos, NEDI, etc you need to go elsewhere. There are plugins to Premier Pro and Photoshop with Lanczos and maybe some others. It is the same algorithm problem.

Both SI2k and RED One are capturing RAW with wavelet compression. The SK2k, because of the sensor isn't oversampled actually can output uncompressed RAW (not sure if they include software to capture it though).

So all the pieces are there...just the funding to put them together. What the RED Scarlet and Canon 5DM2 did is place them front and center in the sub $10k price range. Now it is just time until the manufactuers, under competitve pressure if nothing else, respond.

David, these comments are the essential reflection of this thread and should be the final words of it.

 

[hawaj]

Isn't it just some Canon vaporware or rumor scam?
He he, looks like everybody takes it for granted.
Wonder if it they really show workable prototype at NAB or just some crappy mockup full of air. And Panasonic too. He he.

 

[dadoboy]

Rumor or not, it seems RED's leadership is taking the threat seriously if you read what's going on at reduser.net. Remember, theres Nikon/Canon/Panasonic/Sony plenty of companies to come up with a large sensor aps-c single chip video camera. As nimble as RED may be, it may be hard for it to compete with the likes of Canon on the "low" prosumer end especialy since Canon makes its own glass and chips. RED is getting its chips from manufacturers like Dalsa. My guess would be that RED may end up a higher end player against the likes of Arri, Panavision and Sony. Yes, just rampant speculation!

 

[dcloud]

I dont think RED made that huge impact on the big companies market share.

 

[hawaj]

From where is Red getting their chips is mystery and I think they are pretty sure with their source (maybe they somehow own it) because with these chips all stands and falls.

Every time when competition post some rumor Jim act like that. It looks like he is reacting on new facts, but he is very well prepared in advance and only choose from preplaned moves. He is just playing and having fun with other companies and that is great because competiton is healthy and we profit from it.

 

[SPZ]

Sergio, you were expecting anything more? lol. Come on now.

Jim Jannard just posted a "sort of" commitment to delivery on the End of Summer (pending ASIC delivery/development hurdles).

http://reduser.net/forum/showthread.php?t=28458

Kholi, you know what? I'm excited by all this. Seems like this year, I'm SURE to be getting a camera that, a year ago, was considered nonsense and unatainable for less than 100 000 USD. Thanks to RED, and Canon 5d MarkII.

What can I say? Woohoo!