This is a nice explanation. If you don't mind a suggestion, consider recording some human audio for this or others in the future, I think it'd really improve the impact and digestibility.

Awesome. Seriously, great! Very useful!

So this is how FSR works, extremely cool...

Recommend a program to resize an image with a Lanczos filter (2 and 4)

I did this kind of subpixel interpolation for image rotations of RGB images circa 2000 or so, and I too was using Hungarian math dude's excellent filtering method, but I experimentally adjusted the radius, dampening, and so on to get the lowest error for when the image was rotated back to the original orientation by using Euclidian geometric distance calculation between the before image reference and the after rotating forward and back transformed image--least squares of the differences between (Red1 - Red0)^2 + (Green1 - Green0)^2 + (Blue1 - Blue0)^2 and summing all of those for all the pixels in the entire before and after image rotation. The LS error quantities for larger images were of course HUGE. I also never used 90-degree rotations--why bother when those can be done by transposing pixels from one place to another? I was also considering if I should process separately the R, G, and B, or convert the color space to other models where image brightness was considered most important like in NTSC where the color of the pixels uses less bandwidth in the signal. From what I recall I used a known model for pixel image intensity that had the same response curves for sensitivity that matched the human eye. The rights to the source code were sold to the funders of the imaging library that our group had developed. The reason to do this stuff is semantic image understanding by using complex-valued neural networks that are affine transform immune, then using a lot of training data so that various objects from the real world can be identified, etc. My research was cut short by the tech meltdown where the major contracts for my software development consulting employer were cut off by them having hired too many dead-weight employees doing all kinds of fancy artwork adornments for the image and text pages served. Then at around the same time, the sole company owner died due to a heart attack or stroke by remaining unsupervised in his home sauna while his wife/girlfriend had gone to bed sometime beforehand. His name was John Garay and he died summer 2002. It was not known to me at the time, but I recently learned why it was that so many businesses went under so that their stock value became worthless. Glenn Beck explained how most of the key companies to remain unscathed by this meltdown were being fed capital through shell companies for the three-letter espionage and intelligence organizations of the US Federal Government of the time. So, it was not deep pockets of rich investors doing this cash burn rate funding, but Uncle Sam and the taxpayer. Who would know otherwise? In any case, that's actually unethical, unfair, or unsportsmanlike conduct ("dirty pool" 1940s expression)--fraud--massive market distortion. The beneficiaries were Google, Oracle, and some others who were getting this infusion of tax dollars. This tech meltdown may have been caused by click-through ads on web pages having a huge collapse in pricing. Who is it that does a lot of ads? Who might have artificially priced the ads way lower than the existing market value at the time? Probably the very same people who ended up cornering the market that they collapsed by using these Uncle Sam and taxpayer cash infusions.

I see somewhere mentions the second condition is from -a