With SVP 3 I can set artifacts masking just to the point where I can reduce halo/ghosting and other artifacts where they are gone or with in acceptable limits and still retain a high degree of smoothness.
I suspect when the team is finished with SVP4 in October, it will be as good or much better the SVP 3.
With these settings the CPU is just coasting along at around 11 to 18% and even with 4K footage it's still only at around 54% _ that's with GPU acceleration enabled of course.

Chainik
For the record, I would pay for the full version, what ever the price may be.

Sounds like drug-dealing business

Please, don't panic
SVP 4 is far more advanced than SVP 3.
The only question is how to separate "free" and "full" versions.

Anyway, I do agree that a user controlled option (even a normally hidden one) would be the best solution, but it seems like the main (only?) goal of the new SVP4 is 'user-friendlyness' (hardcode most settings and hide the rest; while also incorperating computationally cheap and common image processing functions) and making it run on other operating systems while maintaining 'good-enough' (good enough for who though?) performance on old/cheap laptops.

While I may not like or support these development priorities, I can at least understand the rationale behind it; both as a means of generating more donations (and increasing the final product's sales among the 'clueless masses') and maybe even as a way to promote the virtues of high-framerate videos and movies to dispell the 'soap-opera effect' myth.

Also, those sharp (high spatial frequency) artifacts can unfortunately not be completely eliminated by any combination of mask settings, while still getting a smoothly interpolated result for the same masked pixels aswell.

The main reasons for the creation of these artifacts in the first place (and why they are not cobsidered fixable bugs) are two-fold:

Firstly, SVP uses block-based matching to determine the relevant motion vectors so that all SSE2 CPUs, back to even the very first Pentium 4 chips that are more than 14 years old, would be able to run it.

Secondly, the 'Aperture Problem' combined with temporal aliasing, requires that SVP's interpolation algorithm relies on some axioms regarding the motion flow vector field, while these axioms are most commonly derived from the 'brightness constancy constraint' and the required 'smoothness' of the motion field.
In your case, the beam of light changed the luma of the pixels 'under' it relatively much more than their chroma values, thereby violating the constant-brightness axiom.

However a bit of good news (even if the final SVP4 version ships without any of the current 'advanced' user controllable algorithm values), if you are prepared to give up smoothness to reduce artifacts in 'problematic' areas such as those, then using SVP3 with some tweaked 'override' parameters (and a more powerfull pc) will probably produce an image that is as artifact free as that which the final version of SVP4 would produce.

exactly