mavericks+titan = faster than red rocket

[Blair S. Paulsen]

La Scaler

La Scaler

As a general rule, when the hero deliverable is 1080/2K, I don't even consider any sharpening or other treatment of R3Ds in RCX.

When I'm ready to render out the HD media, I find a shot or two that has detail I'd really like to hold, then process them using 3 or 4 different downscaling filters. Then I review the footage on at least 2 different monitors (or projectors for features) to judge perceived sharpness and look for moire/aliasing. I get real close, then I get at least 3 screen heights back. Based solely on what my eyes tell me, I choose the best filter for rendering that particular deliverable. This may seem like a lot of extra effort (I certainly don't do it for every project) but IMO it pays off because in only one case did I actually need to do any post sharpening - every other time just picking the right downscaling filter got me there. If you've ever had to run a high quality sharpening pass on anything longer than 30 seconds (sans render farm) I'm pretty sure you'll understand why the time I spend picking the filter is a reasonable trade off ;-).

I'm really looking forward to starting with 5K, 5.5K and 6K Dragon R3Ds for authoring UHD and 4K deliverables. Its my assumption that the impact of the scaling process will manifest quite differently and be more sensitive to viewing metrics than 1080 - particularly perceivable contrast ratio (ambient light and glare can so radically alter this metric that I almost don't care what ratio the manufacturer claims) and distance from the screen.

Cheers - #19

 

[Mark L. Pederson]

OH snap… Are there any more tidbits like that? I've just been going by what RCXp says because each "algorithm" (smoother, smoothest, sharper, sharpest, etc.)… I didn't realized different strokes (algorithms) were better for different folks (outputs).

What is "best" in image fidelity is very subjective - so the best anyone can do making software for image processing is to make a "default" setting(s) that are based on a opinion of what output/export that "most folks" want - and then give you the knobs and buttons that allow you to swing it to the left or swing it to the right if you want.

The first thing I noticed when I did some UHD/4K finish from 5K/4K vs all the HD/2K finishing I've been involved with from 5K/4K material is that when you take out downsample - (or dramatically reduce it - 5K/4K>UHD vs 5K/4K>HD/2K) - sharpening and the debayer choice become MUCH more important.

For the longest time - we have had a CPU debayer and a Red Rocket debayer. What MANY people never understood was that in almost every DI application that supported Red Rocket - you could choose to have the SCALING done by the Red Rocket (super fast) - or - by the CPU/GPU (more options of algorithms = more quality control and choice over artifacts or lack thereof).

So .. .lots of people are lumped scaling with debayer and not understanding they are, in fact, two different knobs you can turn.

Reality is - when you downscaled to 2K/HD - the difference is often very tough for average viewer to see - so, most of time, IMO - letting the Red Rocket do your downscale when finishing in HD was/is fine - BUT - when you don't downscale so far - you just go from 5K (or 6K) to 4K or UHD - the debayer choice - and the scaling choice - and the choice to (or not to) sharpen and the method to sharpen - becomes much more critical.

 

[Blair S. Paulsen]

What Mark is saying is along the lines of my expectations. Looking forward to more exploration in this area, especially with various UHD and 4K displays.

FWIW I would add that rec 709 vs P3 has more impact on scaling prefs that one might imagine for a color space. My assumption is mostly gamma curve effects, though reflected vs transmissive and of course size of screen are usually variables as well.

Cheers - #19

 

[Mark L. Pederson]

FWIW I would add that rec 709 vs P3 has more impact on scaling prefs that one might imagine for a color space.
Cheers - #19

I'm gonna have to mostly disagree here Blair. If you are scaling from RAW to rec 709 vs. RAW to P3 .... I'd argue that your choice of scaling is totally independent of color space choice. I guess it's possible that a certain scaling choice on certain material will sample some colors a certain way when sampling/averaging pixels - but I would't say that's impacting choice of scaling prefs - I just think you might see more to make a better choice.

I actually like to pull out the color and look at black and white when I'm choosing the method of scaling

 

[Timur Civan]

lanczos3 is a downscaling algorithm - it's not a debayer algorithm. After the R3D is debayered - IF you want to downscale to 2K or HD, etc. - many applications let you can choose which algorithm you want to use for the downscale. Lanczos is one such option many apps give you - and there are many hybrid, custom versions of these algorithms as well. But downscale is not debayer.

Ahh, thanks Mark!

 

[Charles_Z]

I'm gonna have to mostly disagree here Blair. If you are scaling from RAW to rec 709 vs. RAW to P3 .... I'd argue that your choice of scaling is totally independent of color space choice. I guess it's possible that a certain scaling choice on certain material will sample some colors a certain way when sampling/averaging pixels - but I would't say that's impacting choice of scaling prefs - I just think you might see more to make a better choice.

I actually like to pull out the color and look at black and white when I'm choosing the method of scaling

If scaling is done properly in linear light, then you're correct that it's irrelevant.

 

[Elliott Balsley]

What MANY people never understood was that in almost every DI application that supported Red Rocket - you could choose to have the SCALING done by the Red Rocket (super fast) - or - by the CPU/GPU (more options of algorithms = more quality control and choice over artifacts or lack thereof).

I didn't realize that was an option. In RedCineX, where can you specify Rocket debayer with CPU scaling?