Noise, sharpness and resolution...

[Blair S. Paulsen]

I had another Alexa shoot today and was able to tell them - with confidence - that they could easily blow up 100% (even though I knew 200% was possible). It really would be worth a proper test sometime. Personally - I hate the idea of reframing and blowing up later - but as a test in how resolution holds up (4K vs. 1080) these recent experiences I have had really make me wonder that math has little to do with it.

Before we declare that math has little to do with these issues i have to point out that the Alexa is doing an internal downsample to create that 1080 image you are blowing up. Because the higher native resolution of the Alexa Bayer pattern CMOS sensor (it's 2,880 by 1,600 in ArriRAW and I believe even that is slightly down sampled from the active photo site count of 3168 x 1782) allowed it to capture more edge detail to begin with, I am not surprised it holds up better when blown up than one might expect. Expanding on Jarred's note, when the material is sharp, clean and has lots of mid frequency detail; blow ups typically play well.

Cheers - #19

 

[Graeme Nattress]

(it's 2,880 by 1,600 in ArriRAW and I believe even that is slightly down sampled from the active photo site count of 3168 x 1782)

It's a crop from the sensor, not a downsample to get the raw. You can't really downsample raw. Well, you can, it's called line-skipping and looks atrocious....

Graeme

 

[Blair S. Paulsen]

Fair enough Graeme, but my point is that even if its a scale from 2,880 by 1,600, rather than 3,168 by 1,782, to the 1,920 by 1,080 output there is still likely to be more captured edge detail in the Alexa 1080 file than from a native 1,920 by 1,080 CMOS chip. Beyond that, many people shooting on the Alexa are using top flight optics and well crafted lighting which tends to improve the results of upscaling as well.

Of course if blow ups are in your plan then Epic over Alexa is an easy choice - that is if you are the one deciding ;-)

Ultimately when someone questions the validity of math I find it hard to stay quiet... ,-)

Cheers - #19

 

[Dylan Macleod CSC]

Ultimately when someone questions the validity of math I find it hard to stay quiet... ,-)

Cheers - #19

...and this is why I was so baffled by the 200% blow up that looked great. I have no math argument to explain what I saw. And that was "before" sharpening - for those who conveniently overlooked that in my previous post.

 

[Graeme Nattress]

Sure, there should be more detail in a 2880 capture downsampled by 2/3 to 1080. When I plot MTFs I see ArriRaw and ArriProres getting 50% at about 1.25k in both cases. Epic was >2.5k at 50% MTF.

Graeme

 

[Antonio Ribeiro]

I think its very smart that you have not released a proxy module. Even now after Red has been around for years, I still see companies immediately rendering their clips to proRes with a redgamma 2/3 curve and using them throughout the pipeline all the way to the finish. The image is good. But, man if they only would do a little research and see that they don't need to do that anymore. I have my own camera with a computer and I edit in raw right up until my final render. I just hope that a proxy module doesn't diminish the efforts that companies like Adobe and Blackmagic are putting into accommodating Raw workflows. And just FYI Speedgrade is fantastic! it was annoying at first but I got used to it pretty quick.

Coming from Color to CS6 Speedgrade was (is) a killer! I just didn't like it and I still don't like it....

 

[Mark Toia]

Shooting 1080P masters makes as much sense as recording to tape...

Recording RAW and simultaneous DNX-HD, ProRes or H264 for editorial makes sense. Recording one of those for your master... doesn't. (rolls eyes)

Let me think. What else doesn't make any sense whatsoever?

Jim

Thanks Jim for the future... Your correct, 4k / 5k / 6k is the furure, we are all thankful for that... and always will be.

But... 80% of us want a current day super speedy 2k workflow as well (sorry) that suites the current day. yes I know, I know... HD sucks... Pro res sucks.. we get that. But alot of people are after quick turn around times with compress HD PRORES or DNxHD files to quickly smack in and out of edit suites.. bang it onto the TV... wham bang done, show me the money. The same as the Alexa boys have.

We will all head to 4k land, there is know stopping that, and with no aurgument. But a quick prores/DNxHD work flow (in a mini module) is what we are all screaming out for. So the rental guys and the TV guys can compete against Arri's illusion.

Don't let VHS be killed by Betamax again... Give us the tools to fight your war.

Oh... forgot to say, I'm personally very happy with my RAW workflow ... But I do master out my finals to PRO RES... It's just easy. I'm sticking with it.

Cheers mate, love ya work.

 

[Tony Lorentzen]

Thanks Jim for the future... Your correct, 4k / 5k / 6k is the furure, we are all thankful for that... and always will be.

But... 80% of us want a current day super speedy 2k workflow as well (sorry) that suites the current day. yes I know, I know... HD sucks... Pro res sucks.. we get that. But alot of people are after quick turn around times with compress HD PRORES or DNxHD files to quickly smack in and out of edit suites.. bang it onto the TV... wham bang done, show me the money. The same as the Alexa boys have.

We will all head to 4k land, there is know stopping that, and with no aurgument. But a quick prores/DNxHD work flow (in a mini module) is what we are all screaming out for. So the rental guys and the TV guys can compete against Arri's illusion.

Don't let VHS be killed by Betamax again... Give us the tools to fight your war.

Oh... forgot to say, I'm personally very happy with my RAW workflow ...

Cheers mate, love ya work.

I'm printing this post and framing it.

 

[PatrickFaith]

... But... 80% of us want a current day super speedy 2k workflow as well (sorry) that suites the current day. yes I know, I know... HD sucks... Pro res sucks.. we get that. But alot of people are after quick turn around times with compress HD PRORES or DNxHD files to quickly smack in and out of edit suites.. bang it onto the TV... wham bang done, show me the money. The same as the Alexa boys have.

I would love it if redcine trim would create 2k r3d file/proxies - 4k to 2k wavelet transforms give a 1 stop increase which would be super nice(i.e. 4 sensors are wavelet merged/added to 4:1, i don't like sampling down btw). I am finding r3d linear to be magical and supported in all the key packages i use. I'm sure a lot of people would love it if we could have 2k r3d wavelet transformed proxies for vfx workflows(also would make the transition to 6k seamless). Lot of crazy stuff we will be able to do with 6k to 2k transforms, like single camera deep pixel transforms(i.e. single camera depth mapped pixels for a 2k openxr 2.0 deep pixel format). So excited to see what you guys do next!

 

[Mark Toia]

I'm printing this post and framing it.

Some may disagree, but I know a majority don't. My rants are probably being filed into the round filing cabinet beneath the RED desk )

 

[Matt Ryan]

Some may disagree, but I know a majority don't. My rants are probably being filed into the round filing cabinet beneath the RED desk )

Mark, your posts are brillant! Seriously I'm so tired of hearing about how easy Arri is and having to sell the Epic extremely hard to get it on a job. A pop it on the back thin prores/dnxhd module is needed, like a year ago....Tethering a pix is not ideal to work with. I love having the RAW to go back to but a quick 2k file to edit and play with is perfect for 80% of the work myself and most do... Plus it would put to rest the idea that you need a Redrocket to work with Red footage.

 

[Adrian Jebef]

We will all head to 4k land, there is know stopping that, and with no aurgument. But a quick prores/DNxHD work flow (in a mini module) is what we are all screaming out for. So the rental guys and the TV guys can compete against Arri's illusion.

Market share isn't an illusion.

 

[Adrian Jebef]

Mark, your posts are brillant! Seriously I'm so tired of hearing about how easy Arri is and having to sell the Epic extremely hard to get it on a job. A pop it on the back thin prores/dnxhd module is needed, like a year ago....Tethering a pix is not ideal to work with. I love having the RAW to go back to but a quick 2k file to edit and play with is perfect for 80% of the work myself and most do... Plus it would put to rest the idea that you need a Redrocket to work with Red footage.

Well, the ALEXA is easy and looks good. So I guess you're gonna keep hearing about it for awhile...

 

[Matt Ryan]

Well, the ALEXA is easy and looks good. So I guess you're gonna keep hearing about it for awhile...

Of course, but making it easier to go from production to post, the less extreme the difference between the two workflow wise.

 

[Karim D. Ghantous]

Let me think. What else doesn't make any sense whatsoever?

Instagram. Shooting super-fast lenses wide-open just because it's fashionable and you can brag about it on your blog. Shooting any serious work with a Canon 5DII (if you have a low budget, you should be using a 7D or equivalent; if you have a high budget, you should be using better cameras). A diesel powered 'supercar'. The food pyramid. Borrowing money to buy a car. Processed milk. The Porsche Cayenne and its stablemates. A lot of medical practice, assumptions and conventions. Etc.

Down-rezzing lowers noise.

Sure, but not so much as you think, if you mean down-sampling to merely half-resolution (e.g. 4K to 2K). A lot of detail and therefore noise is preserved if your delivery file is half the resolution of your acquisition file. We all acknowledge that, all other things being equal, 4K to 2K looks better than straight 2K.

From what I can gather, you need to go to quarter resolution (e.g. 8K to 2K) in order to significantly minimize noise. But I think we all understand the basic point you're making.

 

[Björn Benckert]

Instagram. Shooting super-fast lenses wide-open just because it's fashionable and you can brag about it on your blog. Shooting any serious work with a Canon 5DII (if you have a low budget, you should be using a 7D or equivalent; if you have a high budget, you should be using better cameras). A diesel powered 'supercar'. The food pyramid. Borrowing money to buy a car. Processed milk. The Porsche Cayenne and its stablemates. A lot of medical practice, assumptions and conventions. Etc.


Sure, but not so much as you think, if you mean down-sampling to merely half-resolution (e.g. 4K to 2K). A lot of detail and therefore noise is preserved if your delivery file is half the resolution of your acquisition file. We all acknowledge that, all other things being equal, 4K to 2K looks better than straight 2K.

From what I can gather, you need to go to quarter resolution (e.g. 8K to 2K) in order to significantly minimize noise. But I think we all understand the basic point you're making.

Hmm,
Love where it all goes...6k dragon with a bigger and more pixel dense sensor toghether with our x2 anamorphic hawks.... Can you just imagine the quality of the image when you make a full x2 vertical downsample from that sensor... It makes so much sense. I totally understand the arri approach of their 4:3 sensors and their new MP anamophics. Getting a x2 vertical down sampled image from a CMOS sensor is a great way to get around most of the debayering debate. This we shot with Alexa studio, codec recorder and our hawks C, I think it's the 75mm: http://www.syndicate.se/Files/~usr/hawk/guyinfactory.jpg

click on image to see 1:1


Using the dragon sensor will be a much much better than the above, even using the Epic with MX sensor is better, using the scarlet with 4k is better. I think people will learn how to handle r3d's and that a larger sensor with higher resolution is what to look for. Dealing with codec recorders is not that easy in comparison...

Even though a proress module seams like an easy thing to fix, I find it quite lame to complain on RED for not having any onboard proress module... since I find it much more relevant to have an onboard RAW module... Alexa does not have that, the codec recoreder is pretty darn far from being a small click on module... and it does not record anything more than 30 fps if you do not buy the highspeed version...which can record a wopping 48fps.... And I can tell you as much as from my experience, Im not at all impressed of the uncompressed arri raw files, kind of came like a small suprise but the noise we see in low lights on the MX is not so much compression as we might think it's, since on the Arri raw uncompressed it's just as evil, or I would say worse, since the uncompressed look of this sensor noise is something I find looking very digital where as the 3:1 noise is looking more organic. Did not think that that would be my reaction but testing and viewing results is far different from theory.

Keying on underexposed arri raw uncompressed files is quite a challenge, due to this super high per pixel noise.

 

[David Birdy]

My clients need a 1080i live output signal...has anyone had success with the brighteye product? I'm open to ideas if anyone knows of a good interlace inserter device.

 

[Peter Moretti]

... I think people will learn how to handle r3d's and that a larger sensor with higher resolution is what to look for. Dealing with codec recorders is not that easy in comparison...

There are some workflows for which it makes no sense to use .R3D's... and with their resolution and size increasing, this fact isn't going away. I do agree that the inclusion of a RAW recorder in Red's cameras is UTTERLY AMAZING.

 

[Steven Caesare]

It's a crop from the sensor, not a downsample to get the raw. You can't really downsample raw. Well, you can, it's called line-skipping and looks atrocious....

Graeme

I've wondered about this. Not to diverge too much... but a quick technical question if you'd be kind enough to indulge me, Graeme:

My understanding is that the individual R, G, and B "image chunks" from the original bayer sensor can be stored individually, with the G portion having 2 chunks, as each set of 4 bayer pixels have 2 G's. One could even use a JPEG2000-like wavelet compression routine and store each of those chunks in a compressed format (say in the R,G,B, and A channels), and have a compressed RAW storage codec. :yesnod: Upon decompression you may then debayer by combining all 4 chunks in to an RGB final image.


Seeing as how each "chunk" is simply a monchrome bitmap, what prevents one from scaling each chunk prior to compression/storage? I had considered the lack of co-sited pixels, but it seems to me that as long as your scaling factor was identical for each chunk, you'd still have pixels in the "correct location" for the debayer algorithm in the final downscaled bayer image.

Any insight?

Thanks.

-sc

 

[Graeme Nattress]

Steven - you could do this, and you could try it, but the results are inferior to doing a full demosaic first, then your downsample. Remember we use good optical low pass filtering ahead of the bayer sampling to make for negligible aliasing in the final image. If you did as you suggested you'd also need the software equivalent of that optical low pass filtering too, and although electronic filters can be much more efficient than optical ones, that now you're having filter to a lower frequency still outweighs the benefits of the electronic approach.

Graeme