Rethinking Compression w/RED's 8K Helium Sensor - Pt.1
I'm going to call this post "part 1" because I want to shoot and post more real-world tests with human subjects, low light, high frequency color, etc. - but I felt compelled to share some early test observations.
RED cameras compress RAW using variable wavelet compression. In the early days of RED one we had RC28, RC36 and RC42. By the time we moved onto the Epic and Scarlet DSMC(1) cameras RED transitioned to ratios - and they gave us a whole crazy range to choose from - 3:1 to 18:1. For a point of reference - RC28 = 12:1, RC36 = 9:1, RC42 = 7.5:1
Of course, the obvious concept is less compression is better for image quality - but worse for media management speed and cost.
So what is the "right" compression setting?
When people ask me this question - I always try to explain that it actually depends on what you are shooting. Wavelet compression works with frequency of the image. Think of frequency in an image as the rate of change. Parts of the image that change rapidly from one color to another (e.g. sharp edges) are high frequencies, and parts that change gradually (e.g. large surfaces with solid colors) are low frequencies.
If you turn on the EDGE mode on a RED camera - which I constantly preach as the best focus assist tool you could ever dream of - because it is sub-pixel / pre-debayer - it will show you a visual indication of the amount of high frequency detail in the frame. If you see a lot of white lines ... you have a lot of high frequency detail and you will benefit from a low compression setting. I always thought it would be interesting if RED had a button that would choose a compression setting for you based on this data.
So, a wide shot that is all in focus with high detail, like a landscape of grass and trees will need a lower compression setting to be visually lossless than a medium shot of person with a very shallow depth of field (blurring background with no detail).
When your compression setting is too high for the frequency in the image - you loose detail. Which, as artifacts go, isn't the worst thing in the world. Better than "blocks" or "jaggies". Often, a shot on RED with too much detail for the compression just looks like the AC borked the focus.
Now, in the real-world - we are not going to want to think about compression on a shot-by-shot basis. "yeah ... this looks like a 6:1 shot, etc." That would be crazy. The solution is pick a compression setting for the job that is essentially, over-kill.
For a long, long time - 5:1 has ruled as the "feature film" / "high-end" compression setting. Mostly because only the rarest shots benefit from 4:1 or 3:1 over 5:1 - and 4:1 or 3:1 is much harder and slower to manage and transcode.
So ... most of us got lazy and just shot 5:1 for everything and "slummed" down to 8:1 or 9:1 for a TV promo. At OFFHOLLYWOOD we supported countless feature films at 5:1 with budgets from $100,000 to $30,000,000.
Flash forward to now. RED has released a new sensor - the 8K Helium. And aside from the fact that you have more resolution to down/over sample - it's a different sensor design with a different pixel-pitch, etc.
Which means ... we need to re-think and test.
So here's a rather telling test IMO.
6K Dragon 5:1 vs. 8K Helium 10:1 - both scaled to UHD (3840 x 2160) using new IPP2 debayer (yes, it's a new, improved debayer that is sharper).
No post production sharpening. Just export to TIFF form RCX and scaled in Photoshop to 3840 x 2160.
Pixel peep this a bit and you will see ..
10:1 8K Helium > UHD does in fact resolve more detail than 5:1 Dragon > UHD
Let all the ramifications of that sink in ...
When you factor everything in - shooting 8K today is cheaper and faster than 4K, 5K, 6K was just last year. This is something I will keep re-enforcing with more data and real-world examples.