What exactly do "sharper" and "softer" do?

I ask this question purely out of curiosity, but what do the "sharper" and "softer" filtering options do in the edit timeline? I've noticed the "sharper" option causes rippling artifacts around distinct edges and it screams FFT to me, but beyond that, I have no clue how it's processing the image.

Just curious if the community has any ideas! The documentation just says that they're sharper/softer scaling methods proprietary to Resolve, which... duh.

I ask this question purely out of curiosity, but what do the "sharper" and "softer" filtering options do in the edit timeline? I've noticed the "sharper" option causes rippling artifacts around distinct edges and it screams FFT to me, but beyond that, I have no clue how it's processing the image.

Just curious if the community has any ideas! The documentation just says that they're sharper/softer scaling methods proprietary to Resolve, which... duh.

https://www.bilibili.com/video/BV1784y1N7YE/
If you ask the filter, you can look this video.

Scaling filter is a weights kernel that is used when resampling image. Sharpening kernels have negative weights in some components which exaggerate edges but can cause artifacts. Softening kernels don’t have negative weights, so they average source pixels together, causing slight blurring.

Whether fft is involved in scaling… don’t know. Since there are other transforms at play too, that can be applied, I doubt it.

Scaling filter is a weights kernel that is used when resampling image. Sharpening kernels have negative weights in some components which exaggerate edges but can cause artifacts. Softening kernels don’t have negative weights, so they average source pixels together, causing slight blurring.

Whether fft is involved in scaling… don’t know. Since there are other transforms at play too, that can be applied, I doubt it.

Oh, interesting! I thought it was strange that "Gaussian" was an option for scaling but it didn't occur to me that convolution kernels were also used for scaling methods. The visualizations for scaling are always bilinear/bicubic filtering and are shown similarly to MSAA visualizations but with points and distances or with a waveform view.

The math itself goes over my head, but I like to have at least an intuitive understanding of what's going on. I'll update this reply with any good visualizations I find.

Thanks