Realistic compression for the next hdslrs

arrestthisman

Well-known member
Hello all, I know this has been talked about some, but I haven't found a thread that really dives deep into what we can realistically expect from the next generation of video from these wonderful new HDSLR cameras.

There has been rumor of a Raw recording module from Canon. While that would be a dream come true for filmmakers, the storage and speed required for this at a sustained rate makes it an unlikely endeavor. Even with sRAW, unless Canon employs an amount of compression previously unheard of in their RAW format, I just don't see it happening. That said, I hope it DOES happen, but to the point. Realism...

I began thinking to myself, after taking a series of 8mpix JPEGs for some time-lapse, that the footage after downscaling to 1080p looked far "cleaner" and sharper than most 1080p camcorders. Well, of course, we're talking .8 to 1.3mb per frame. So even at 8-bit 4:2:0, MJpeg at something like 1mb per-frame would look great, and is within the ability of CF cards. 24mb/sec for 24p. Megabyte, not megabit.

Then I began thinking about Jpeg 2000. Or I guess Motion-jpeg2000. And did a little reading on it. You'd get 10-bit, and 4:2:2 (I think you can get 4:4:4 too). (I also think you can do better than 8-bit on regular JPEGs.)

These are just my thoughts, but I bet the seasoned out there would know much better. That's what I'm interested in hearing.

I think the idea is, though, RAW is great, but my experience is that extremely low compression yields footage almost as mailable as RAW. And I've found uncompressed "baked in" footage to hold up better in CC than compressed RAW.

Plus, Canon originally put the video feature in there for photojournalists. I don't think they want to give up that segment, and I doubt journalists have the time for processing RAW in motion. On the other hand, if you can add a compression scheme that works well-enough with small bitrates, but can scale it up to master quality bandwidth (even if it isn't the BEST codec) with a simple setting in the camera, you can please both filmmakers and journalists without the need for two separate products.

Lastly, I don't think it matters how cumbersome the video is. Look at the hoops filmmakers have jumped through to get the "look" from these cameras.

With the 7d, Canon upped the bitrate, not a ton, but some. Maybe that's what we are to expect? Nothing radical, but simply as much bandwidth dedicated to video as possible.

Lastly, if they could simply get an image sequence of Jpegs (even without a container) from a camera running 24fps that would be great. What would be even better is if they could give us much more than 1080p with those. Like my old rebel, which is 8 megapixels. If I could get 24 Jpegs/sec from that*, which is 3800 x 2300 or something like that. I could transcode that nicely into pro res 4444, or even just HQ, and I bet the quality would be amazing.

*obviously I don't expect Canon to make my rebel do this. I mean in a new product, which would probably have many more pixels, but if they could get between 5 and 12 megapixels in a simple image sequence, that would do a lot for filmmakers, I think.

Alright, so am I insane, on to something, trivial, or redundant?
 
I dont really get compression

but the nikon D3 (11mp) even shot on crop sensor (6mp) at jpg low creates files that blow virtually any video I have seen away

considering the (D3) cam can already burst that at 13FPS it would seem no major acheivement of engineering to up that to 24FPS and output it to some device with a HD

Seems simple to me - better than pixel binning etc

S
 
Depending on how long you think the "next generation" will be from now (I'd say 6 mo), I'm looking forward to REDCODE and CinemaDNG.
 
but the nikon D3 (11mp) even shot on crop sensor (6mp) at jpg low creates files that blow virtually any video I have seen away
Part of that is probably due to the still mode using a mechanical shutter. Mechanical shutters are used on DSLRs because they offer significantly improved image quality over current electronic shutters. So the difference in image quality isn't just due to pixel-binning and compression.

Remember that the shutters are rated for 150,000 to 200,000 actuations. At 24 fps, that's only around two hours of video.
 
Well, the Casio EX-F1 is not a DSLR but you can do lots of crazy tests with it. Can't remember how big is the buffer but I read some people could do a little trick with it for timelaps and if you cranck the fps might be good for video (using jpg sequence, I know it already shoots usless 1080i but very nice 720p).
 
But you have to take aliasing and line skipping into consideration its how they downscale ,not like you did with your jpegs and probably lanczos3 algorithm.
First let them downscale properly in next cams.
 
First let them downscale properly in next cams.

+1

Looking at footage from my 7D, I say that aliasing/moire due to low quality downscaling from 18mp to HD rez is way more detrimental to image quality than compression artifacts.
 
Lastly, if they could simply get an image sequence of Jpegs (even without a container) from a camera running 24fps that would be great. What would be even better is if they could give us much more than 1080p with those. Like my old rebel, which is 8 megapixels. If I could get 24 Jpegs/sec from that*, which is 3800 x 2300 or something like that. I could transcode that nicely into pro res 4444, or even just HQ, and I bet the quality would be amazing.

This para sums up the most important part of post one. Forget video mode, essentially a continuous frame burst at 24 jpegs/sec.
Aliasing goes out the window because there is no line skipping or pixel binning.

I don't think RED would go down this road but a smaller competitor (ikonoscop link or similar) might.

@OP - I like the way you are thinking
 
This para sums up the most important part of post one. Forget video mode, essentially a continuous frame burst at 24 jpegs/sec.
Aliasing goes out the window because there is no line skipping or pixel binning.

I don't think RED would go down this road but a smaller competitor (ikonoscop link or similar) might.

@OP - I like the way you are thinking

Well the cool thing about a jpeg sequence is that it is not very cumbersome. You can bring it into any space you want. It's oversampled, but not in a proprietary format. Up the buffer in a cam, or just the speed of CF cards, and you can lower the compression ratio as the tech improves.

The RED heads would look down their nose at this, but who cares, it would still be a big step up. Then, once CF cards that can write 240MB/sec for uncompressed, and 2TB drives cost a song then you can go into raw or just uncompressed.

Also, for audio, I don't think it would be that hard to record PCM at the same time through a minijack (with the intention of using a beachtek or something for XLR). The PCM file could be given a serial number that matches a portion of the image sequence file names (or just the folder the image sequence records in. And then no headaches--- auto-matched audio without headaches, and without the need of a wrapper like mov.

4:2:0 is meaningless, also as long as you oversample it to the point where you bring it into a 4:2:2 or 4:4:4 container at 1/4 the size. Besides, who needs a native 444 codec when you're doing bayer!?!

Plus, even huge jpegs can display excellent quality at 1mb per frame. This also means that as CF speeds improve (as long as the sensor can keep up) then overcrank is limited only by speed.

An issue for pro use could be no monitor out, but with an actual video DSP, you could probably setup a feed to HDMI for critical focus. Also, we're talking about very low budget, so if they could get a binned 1080p out to a monitor for critical focus even just when the camera is in standby, you could do a run-through and get your marks for follow focus, and record blind... Peter Jackson shot blind with the RED, no reason it can't be worked around for the purposes we are talking about.

I also think that more SI-2k-mini type systems would be great. I'd love to bring a laptop with a tether to a high quality head, but I think SI-2k is way overpriced with all of the other options. That said, the workflow is supposed to be awesome. But here's a thought:

If all the stuff I'm talking about is not possible in-cam above, with the jpeg streams. What about this:

Make an HDSLR just like now. Please the still-shutterbugs, photojournalists, web-video, etc. Then add a gigabit ethernet out, and commission software (I bet money that some filmmaker/programmers would do this for free! let alone payment in the form of a camera) that captures the feed from the cam, and have the laptop do the work.

This way, critical focus is taken care of. Recording is limited to the computer capacity. (A G-raid mini would make short work of any master quality system), and surely it could handle a simple jpeg/wav stream. To make it even easier on Canon or Panny or whoever would look into this model. Don't support it. Just provide as many tools for programmers as possible. And if they don't want their trade secrets out, add a general purpose DSP that programmers can do whatever they want with.

Open source camera creation! You could also have applications where productions could pay programmers to modify or "tune" these cams for their specific needs.

Any company that approached the filmmaker's target market this way would be a hero to me.

Some cool stuff to think about.
 
Make an HDSLR just like now. Please the still-shutterbugs, photojournalists, web-video, etc. Then add a gigabit ethernet out, and commission software (I bet money that some filmmaker/programmers would do this for free! let alone payment in the form of a camera) that captures the feed from the cam, and have the laptop do the work.

Great idea. I'd be amazed if they ever did such a thing. I'm sure some hardware guy will explain why it could technically never happen either, but thats just negative nelly talk.
 
Great idea. I'd be amazed if they ever did such a thing. I'm sure some hardware guy will explain why it could technically never happen either, but thats just negative nelly talk.

True, but it's been done in a way... Andromeda. From the limited knowledge I have about video engineering, andromeda should have been a larger challenge than anything I'm talking about here.

Maybe we should organize something.

Does anyone here know anything about where to start to lobby a company to do something like this?
 
Back
Top