First off - sorry about ‘Button 1’ - I made this on a whim, that’s a temporary testing button
I just tested out my project on an Amazon Mac M1 (for the hell of it) that’s 3,600 miles away from me - the screen refresh rate is not exactly representative as a result (I think it’s measures in seconds per frame rather than fps)…
Oh, and AWS’s Mac M1s only have a screen res of 1024x768 so that don’t help either
This is Python4Delphi with PythonEnvironments, Lightweight Python Wrappers and P4D Data Sciences all helped along with a little Skia shader magic.
The front end is Delphi FMX tied to a Python backend that performs image styling (and training) usiing Fast Neural Styling via PyTorch.
The results are rather slow ATM as I’ve yet to enable the M1’s GPU (simple enough to do).
It looks a lot better if the machine being used is in front of you My Windows Laptop does the same thing almost instantly (I’ve got a 3060 in it which helps a lot). Enabling the M1’s GPU will make the above take far less time - I’d guess it’ll be under a second (it’s a crappy little Mac Mini - not exactly a powerhouse).
A prevoius TRAINING test revealed that a Mac takes 10 hours to produce a style while my 3060 takes 1 hour 45 mins and a 3080 (actually an A10G - which is about the same speed, possibly faster) takes 45 mins.
CPU-only my laptop takes FOREVER (so the Mac’s actually got it’s good points) - not exactly timed training on a CPU 'cos it takes so long but c 30 hours…