resolution vs redcode trade-off

[George Butterfield]

I shoot 4K HD at RC7.1 I think that's more than adequate for my target, the web and broadcast and Blu-ray (might try RC8.1). I crop in pre.

 

[Mark L. Pederson]

Motion creates new blobs of incorrect macroblock estimation on every frame, so the macroblocking cannot be hidden.

The advantage of wavelet compression is that the wavelet algorithm does not divide image into blocks, but analyzes the whole image.

 

[Les Dittert]

Motion creates new blobs of incorrect macroblock estimation on every frame, so the macroblocking cannot be hidden.

Robert, so true. Our brain's visual processing is very motion sensitive ( evolution set that, as things that will eat us usually move ! )
So the compression artifacts they remain still don't attract attention.

 

[Blair S. Paulsen]

Mark is correct. Wavelet compression uses an entropy encoding scheme that is considered desirable because, for the most part, as it runs out of data it just gets softer.

I believe this same property can lead to the "plasticky" skin issue. I haven't done the test myself, but in theory, if you did a talking head interview in front of a flat cyc at X:1 compression, then changed the background to something busy/moving - but everything else the same - the skin rendition would change. I think of it as there being only so much of the signal devoted to high frequency detail and if its occupied encoding all that other stuff there's less left for subtle detail in skin tones.

If someone would like to jump in and provide a more specific and accurate description please do.

FWIW when shooting overcrank I usually have better results with fewer bigger frames and less compression, using Twixtor to hit my target frame rate in post. As Bjorn notes, not all slo-mo can be treated the same way, there are cases where a bit more noise is a fair trade for more real frames. If I have the time and resources I do just what Bjorn suggests, shoot the same thing at least two different ways.

Cheers - #19

 

[Burke Doeren]

Mike, you could always rent a 512GB High Speed REDMag SSD (one of the red-colored drives) to help reduce your compression settings at high speed.

Does anyone know what the lowest RC setting is that you can choose with one of those drives in 5KFF @96fps?

 

[Les Dittert]

I believe this same property can lead to the "plasticky" skin issue. I haven't done the test myself, but in theory, if you did a talking head interview in front of a flat cyc at X:1 compression, then changed the background to something busy/moving - but everything else the same - the skin rendition would change. I think of it as there being only so much of the signal devoted to high frequency detail and if its occupied encoding all that other stuff there's less left for subtle detail in skin tones.

If someone would like to jump in and provide a more specific and accurate description please do.

I did a test where I filmed a busy scene twice. Once normal, and again with a card with a hole punch in it placed in front of the lens, resulting in only a small center part of the image in focus. Over 75% of the image was thus devoid of high frequency details.
I did not see a difference in the quality of the center part, which is counter intuitive. Perhaps my base compression was set to too high a quality for the middle to suffer much. Worthy of a retest.

 

[Blair S. Paulsen]

I think the difference would be subtle in the situation you describe Les, especially if the imagery you are comparing includes a lot of movement which confuses the eye. I also think it would require comparing full decodes on a reference monitor with at least a 10 bit signal path, etc.

IAC, among the many changes wrought by Dragon is that the effects of various compression ratios is markedly different than MX...

Cheers - #19

 

[Les Dittert]

compression quality

compression quality

There was no movement in the test, and I was looking pretty close doing A/B wipes at 200% in a compositor. What I didn't have in the test was a section of almost black areas, where R3D's tend to jettison fine details the most. I learned this early on by comparing a RAW image from a 12 megapixel Canon DSLR and the same scene from an MX sensor. The MX totally smoothed out the dark areas even at best quality compression ratios.
But even more interestingly : When I compressed the DSLR RAW derived image of the same scene as a 16 bit Jpeg 2000 in photoshop to the same spacial and byte size as a single R3D frame, the detail level was more preserved in the Adobe compressed frame.
This is good news, in a way, as I also believe that the compression in the Dragon ( as you also stated ) is a higher quality one. So more detail can be preserved with the same size, or as people have experienced the RC number can be raised up and still get good results.

I will explore this more when I get the Dragon camera in my hands. Basically I will do some more compression tests : MX processing pipe vs Dragon. So the image quality from the new camera can be attributed to both sensor AND compression refinements.

 

[Robert Ruffo New]

Sorry Robert - but this statement is not technically correct.

You are NOT "zooming in on noise" - you are cropping the sensor. With respct to lenses - other the the very high end glass like Lecias and Masters - fall off occurs at the edges of the glass. On most lenses - cropping to 4K gives you an overall sharper image if you sample detail across the entire image. I have ZERO idea where are are coming up with "you're losing at least half a stop in the noise floor if you shoot at 4K instead of 5K" - please elaborate as that is not my experience at all.

Compression doesn't really know the difference between noise and detail - so ... if you are fighting the noise floor - and you go for higher compression - you are more likely to take a hit on the detail you want to protect. This is why I always recomend LESS compression in very low light - and there are times when having less compression will trump more resolution. Depends on how much detail in the frame, how much noise - and what compression settings we are talking about. 12:1 vs 8:1 is a bigger issue than 5:1 vs. 3:1

Noise floor is perceptual/subjective and kind of a soft target (as we all know). The more of the sensor you are using, the more zoomed out you are from individual pixels and individual points of noise - the more you are super-sampling a poor rendition of light reading, due to a lack of light, and getting an result averaged form more info, and is more likely to be acceptable. To give an extreme example, 1K Red footage looks very noisy very fast, so in practice (maybe not exactly technically on a 1:1 basis, but once fitted to fill a screen) to shoot at 1K you better meter for 320 ISO, not even 500 ISO, whereas at 5K 1000 ISO often looks fine. When you display the entire higher res image, and back yourself up to comfortable viewing distance, each point of noise will be less visible, less offensive - and so really you can push blacks more.

It is therefore technically correct to say that the larger your crop of the sensor, with more pixels in it and therefore with each source pixel appearing smaller in the output, the more stops of noise in the floor you can get away with digging into, without disturbing your audience/clients.

As for lens cropping sharper and nicer are two different things. I do not share the resolution and sharpness obsession, and do not feel that sharper always looks better, nor that it is more important than everything else, all the time. 50mm at 1.4 is not the same as 35mm at 1.4, (my math is off here, the field of view difference between 6K and 4K is a bit less than that) as the diaphragm hole on the longer lens is physically bigger at 1.4 than the shorter lens at 1.4. In fact, all aspects of lens rendition change, in a way that I personally like better when you crop to a larger piece of the sensor, although sometimes it depends on the brand of lens, for sure.

As for compression aI 100% agree with you - and agree it depends on the scene and how you plan to grade it, and yes it can sometimes trump resolution. Again, I misunderstood what your post was referencing Thats why I added an OOPS later.