Hello @garagecoder ,
Not sure if you are still around, but I’ve issues with your command gcd_solve_kernel that is used in various parts of your file andy_kelday.gmic, in particular by your command gcd_upscale_tile2x.
When I try :
$ gmic sp bottles +gcd_upscale_tile2x
I get a memory allocation error (like Failed to allocate memory (79.0 Gio) for image (102969,102969,1,1). )
It seems that this command uses invert to compute the Moore-Pensore pseudoinverse of non-square matrix (image named patches) having size 25x102969, which is apparently a matrix that would have size 102969x102969, hence the allocation problem.
Now, I suppose this code was working before, and in a recent release, I admit I’ve changed the code of command invert which now relies on QR decomposition (rather than LU).
Not sure what invert was returning before…
Could you explain what is the purpose of computing invert on this matrix?
I’d be happy if we can “fix” this code, because your image upscale method based on your solver was really good, if I remember well!
Thanks!