r13

[HolyWu]

• All planes of floating point mask are always in the 0-1 range now.
• Remove TCannyCL filter.
• Add AVX512 and remove AVX.
• Migrate to VS API4.
• Add Kroon, Kirsch and FDoG operators.
• Rename gmmax parameter to scale and change its default to 1.0.

---

This discussion was created from the release r13.

[HolyWu]

Some interesting speed comparison of Kirsch and FDoG operators between TCanny and Convolution+Expr implementation using VS R57, core.num_threads=1 and only luma plane of a 1080p clip.

Kirsch
TCanny(sigma=0, mode=1, op=5): 136.35 fps
vardefunc.mask.Kirsch().get_mask(): 73.74 fps

FDoG
TCanny(sigma=0, mode=1, op=6, scale=0.5): 122.13 fps
vardefunc.mask.FDOG().get_mask(): 3.99 fps