Super scale (was Super resize)

This new feature is puzzling me.
What is the very nature of the Super resize algorithm?
Is is some kind of spline resize or is it more likely a super resolution one?
Still dont understand that...

Where is it?

I think it is in clip attribute.

From manual:

"For instances when you need higher‐quality upscaling than the standard Resize Filters allow, you can now enable one of three “Super Scale” options in the Video panel of the Clip Attributes window for one or more selected clips. The Super Scale pop‐up menu provides three options of 2x, 3x, and 4x, as well as Sharpness and Noise Reduction options to tune the quality of the scaled result. Note that all of the Super Scale parameters are in fixed increments; you cannot apply Super Scale in variable amounts.
Super Scale options in the Video panel of the Clip Attributes
Chapter – 1 General Improvements 1-9
Selecting one of these options enables DaVinci Resolve to use advanced algorithms to improve the appearance of image detail when enlarging clips by a significant amount, such as when editing SD archival media into a UHD timeline, or when you find it necessary to enlarge a clip past its native resolution in order to create a closeup.
This is an extremely processor‐intensive operation, so be aware that turning this on will likely prevent real‐time playback. One way to get around this is to create a string‐out of all of the source media you’ll need to enlarge at high‐quality, turn on Super Scale for all of them, and then render that timeline as individual clips, while turning on the “Render at source resolution” and “Filename uses > Source Name” options."

Yes, my fault. The correct name is Super scale, not Super resize.

I had a read to the manual but it's not clear if the Super scale algorithm operates on single frames or if it leverages on Super resolution algorithms that feed detail informations from surrounding frames with optical flow support.

Hence my question, as i'm greedy looking for some Super resolution "solution" directly into Resolve.

Testing 15_b3, just realized that the 'Super Scale' algorithms can also be applied when upscaling clips via the Media Management > Transcode function. Again it just requires setting the Super Scale options in the source Clip Attributes. Only tested HD > UHD thus far, but I'm quite impressed. Depends on the quality of the source of course, but (so called) 'super-resolution' methods I've tried in the past were more prone to line thickening, 'ringing' or aliasing artifacts. Transcode render times are much, much slower than with 'standard' Resolve up-scaling mind you. Not something I'd use routinely, but useful to have in reserve.

.......it's not clear if the Super scale algorithm operates on single frames or if it leverages on Super resolution algorithms that feed detail informations from surrounding frames with optical flow support.

Yes would be interesting to know more about the algorithms employed, rather than speculating.

Testing 15_b3, just realized that the 'Super Scale' algorithms can also be applied when upscaling clips via the Media Management > Transcode function. Again it just requires setting the Super Scale options in the source Clip Attributes. Only tested HD > UHD thus far, but I'm quite impressed. Depends on the quality of the source of course, but (so called) 'super-resolution' methods I've tried in the past were more prone to line thickening, 'ringing' or aliasing artifacts.

Revisiting/revamping some of the methods I've used before, I've modified my opinion somewhat. For clean, high quality HD sources that warrant little or no denoising and (contra-)sharpening, AVISynth routines utilizing NNEDI3 (nnedi3_rpow2) as the doubling-resizer produce 'superior' results (higher reproduction fidelity) as confirmed by quality metrics, and are faster.

[Peter Chamberlain]

Thanks for feedback.
And for clips that are not clean, high quality HD sources that warrant little or no denoising and (contra-)sharpening, how does it compare?

BM should talk to guy who made madVR renderer.
His upscaling technique is quite amazing (even compared to NNEDI) and it's all on GPU which is good for Resolve.
https://forum.doom9.org/showthread.php?t=146228

It's called NGU. This is what Resolve needs.

Thanks for feedback.
And for clips that are not clean, high quality HD sources that warrant little or no denoising and (contra-)sharpening, how does it compare?

I'm looking at that

Thanks for feedback.
And for clips that are not clean, high quality HD sources that warrant little or no denoising and (contra-)sharpening, how does it compare?

I'm looking at that

Hey Bryan, any updates on your tests?