Hardy- wrote:Hello,
as we know, we can set a threshold below which blurry contours remain blurry, and above which sharp contours get sharper.
Is there a method in Resolve that provides two thresholds?
C: Very sharp video material: Do not sharpen this
---- Threshold 2 ---- (my wish)
B: Slightly unsharp video material that needs to be sharpened
---- Threshold 1 ---- (the usual one)
A: Blurry video material: Do not sharpen this
Well, you have to specify which footage and or counters, or edges need to be changed before you try to change them with some tool. This selection process either has to be done manually by user or by some kind of specially trained AI method. To my knowlage, other than very simple luminescence values there is no real contextual way to do it, other than manually selecting specific footage for specific processing or by use of trained AI.
Samurai tool for Digital Anarchy, a third party sharpening plug in is not AI , but has the ability to use edge masks based on luminescence values. Meaning, much like so called smart sharpen tool in Photoshop it can not sharpening the highlights on edges and do instead only work on shadows, or the other way around. But this is pretty basic and broad for precise selection.
Another method could be with use of frequency separation based on details. There are some tools that can do that, but its probably not best for what you want. They can be effective sharpening tools, but require careful use, not a batch processing tool.
Bryan Ray did a tutorial a while back DeSharpen for Fusion to deal with white edge nasty halos from compressed and processed footage when preparing them for green screen keying.
https://bryanray.name/2018/06/18/desharpen-for-fusion/Again, manual process and involves a lot of work.
Maybe there is some clever way in Fusion to set up what you want, but I'm not sure... maybe some scripting and expressions would be needed. I haven't tried it with stock tool.
But I suppose the problem is even if you use frequency separation to apply or not apply additional sharping based on details , there is no way, outside of manual intervention or AI training to know any context of the image. Because sometimes you may want sharp edges, but not sharp skin details, or you want more skin texture but not contours on the edges.
There are some really good sharpening tools I use that come as DCTL's, but they can't really do what you ask. Interesting challenge. if you manually select clips and footage and you know what you want, than its doable, if you on the other hand require batch auto processing than its a different problem altogether. And even AI can be unreliable. And slow.
For example. Topaz Video AI has feature to auto sharpen or not with bunch of parameters the video footage and I think it tries to either do one time estimation per clip, frame by frame or offers suggestion for manual settings. Great. Problem is that its not very reliable. So sometimes you get what you want, and sometimes you don't get what you want, and human judgment is very much needed on clip by clip basis if you want to do it right.
While not perfect, usually my go to sharpening tool in Fusion and Resolve is this.
The following fuse is based on the Contrast Adaptive Sharpening algorithm of the AMD FidelityFX project for developers.
CAS adjusts the amount of sharpening per pixel to target an even level of sharpness across the image. Areas of the input image that are already sharp are sharpened less, while areas that lack detail are sharpened more. This allows for higher overall natural visual sharpness with fewer artifacts.
https://gpuopen.com/fidelityfx-cas/"As I wanted the code to be based on Quality instead of Performance, I applied a contrast detection based on Rec709 Luminance instead of just focusing on the green channel. You can also change the color sliders in case of sharpening should be weighed differently. Furthermore, the original algorithm scans the left/right and top/down neighbors. I also included the diagonal neighbor pixels for better quality. The area of neighborhood pixels can be extended by the Shift control to sharpen lower frequencies but that might lead to color artifacts if pushed too far. "
https://www.steakunderwater.com/wesuckl ... php?t=6107
- Honeyview_CAS.jpg (366.54 KiB) Viewed 305 times
...................
P.S. If you use it with compressed footage, on higher settings it will also sharping chroma and artifacts will be shown. Pro tip. Use blend mode of luminosity to limit the sharpening only the to luminosity not color where you run a risk of making those nasty compression artifacts really viable.
If you use it in Fusion make sure your DCTL node is use with merge node, since merge node offers blend modes. And in newer version of Resolve in the color page, you can right click on any node now and chose composting mode or blending mode of luminosity .
Also in fusion there are other methods that can be used to take it even further, but basic set up usually works well.
Mon May 13, 2024 5:29 am