Needed: RAW ramp for benchmarking

[stevefal]

I have been running a set of workflow benchmarks in order to determine which best preserve the fidelity of original ML raw footage. My personal goal is to edit in high quality without proxies. In order to accomplish this, the workflow needs to:

-  generate clean, high DR clips that are playable in real-time
-  recover, correct, grade footage in high fidelity, in-place while editing (with GPU)

I use Adobe Premiere Pro CS6 and I feel that this is attainable, as I have already come close.

However, the process is still mysterious in that every possible codec and plug-in effect is essentially a black box that has untold capabilities and flaws. In the worst cases, components claim to do one thing but actually do another if you look at output quantitatively.

The best looking output I've achieved so far is MLRAW-> DNG -> AE ACR -> Animation format -> PPro. Using synthetic high bit-depth gradients, I have confirmed quantitatively that the Animation format preserves bit-depth within PPro better than any other format I've tested so far. Mind that specific settings while rendering in AE are critical.

This benhmark allows me to test the actual impact of certain effects. In some cases PPro level controls are operating in 8-bit even though the overall effect is advertised as 32-bit. This discovery was disheartening but I'm glad I figured it out. The result will be a personal workflow that favors certain effects for certain tasks, and avoids others at all cost.

This is all fine, but since my workflow starts in ACR, the DNG -> ACR -> AE process is still a black box, with unknown impact on bit depth. Note that when using ACR with photoshop, you are given the option to open the resulting image as 8 or 16 bit. No such option is offered when using ACR with AE. So what IS the bit depth?

In order to measure this remaining part of the workflow, I would need to have an original MLRAW or DNG image with known bit-depth and dynamic range. With such an image I can use the same process I am currently using in PPro to determine end-to-end bit depth. I can explain that technique if anyone is interested.

So here is the question. Is anyone able to generate a fake MLRAW or DNG image that is, say 12-bit depth and a simple horizontal grayscale ramp? Using this source frame, a wide variety of black box solutions can be tested to see what they actually do to dynamic range:

- Adobe ACR -> After Effects
- OSX Raw2DNG Prores output
- Adobe PPro CC CinemaDNG support
- GingerHDR DNG support
- Cineform convertor
- etc, etc.

TLDR; I need a 14/16 DNG file with a 12-bit ramp that looks like this:

[dude]

Which formats did you test so far?

[stevefal]

Animation Lossless (million)
ProRes HQ (trillion)
ProRes HQ (million)
Avid 444 10-bit (trillion)
Avid 444 10-bit (million)
Cineform 16-bit

I have not tested the AE path completely methodically yet, because I want to start from RAW/DNG via ACR. If DR is lost at that step, it doesn't matter if the output format is high DR. It fact it would be wasteful to render 8-bit content into a 16-bit format (assuming no effects).

[dmilligan]

I bet that AE simply uses the same ACR bit depth as that of the AE project. Why do you need a special gradient to test this? Why don't you just try it with a real image that has a large DR. All you need to find out is basically that it's smart enough to use 16 bit from ACR when it should (when your project is 16 bit) or it isn't and stupidly converts down to 8 bit and then back to 16. I doubt Adobe would do anything that stupid. Seems to me that's the only question you need to answer.

[stevefal]

ACR>AE is only one of many flows that start with MLRAW/DNG. A horizontal ramp is useful because non-linearities and thresholds can be easily visualized using a waveform monitor. And banding is very clear to see visually.

As to what ACR does, I don't know, and therefore want to test. I have found several cases where Adobe does not handle bit-depth the way you'd expect. This is about separating what I'd like to assume from what's really happening.

[dmilligan]

Yes, but once you get past ACR>AE step you don't need fake dng sensor data. From what you stated I was under the impression that you've already characterized everything besides ACR and it was really the only variable remaining.

[stevefal]

As I wrote, there are multiple flows that start with DNG. These do not involved ACR:

- OSX Raw2DNG Prores output
- Adobe PPro CC CinemaDNG support
- GingerHDR DNG support
- Cineform convertor

[baldavenger]

Was anyone able to acquire a grayscale ramp DNG file?  Would be useful for testing the raw controls in Davinci Resolve, and in particular the many luts that are out now.