John Hayes:
I agree that BXT is not performing any sort of mathematical deconvolution; however, the net result is exactly the same and the term "deconvolution" has always referred to anything that undoes the convolution process that blurs images. And, that is exactly what Russ's algorithm is doing. So, I'm comfortable calling it a deconvolution process.
Thanks for the update, John. Very interesting. I am curious about it not being generative AI... I guess definitions there can vary. I dont know how I would define it. Technically though, as I understand currently what he is doing, BXT takes in one image, and generates another with a lower error factor. Generates being the operative word. I don't think it's the same origin image with corrections, it's an entirely new image?
Anyway... I guess I have some trouble calling this deconvolution, simply as a matter of not conflating how all prior deconvolution algorithms work, vs. how a NN processes input and generates output. The results are...SIMILAR. But, are they really the same? I've never seen any classic deconvolution algorithm produce results anywhere close to what BXT can do...either from a resulting quality/error free (i.e. No ringing artifacts!!) standpoint or from a pure speed standpoint. BXT is insanely fast for the quality it produces, clearly not brute force iterative. Classic deconv algos are often very slow, and to get anywhere close to the shrinkage and profile improvements that BXT delivers usually requires significant iteration counts.
Similar results technically, but BXT is vastly superior with generally error free results. So...I dunno, seems unfair to call it just another deconv algo. 🤷♂️
