Topic: Search radius quality

Hello everyone,

Does anybody know if there is a way to improve the search radius quality? For example have "Average" and "Large" better detect artifacts at the cost of performance?

Thank you in advance.

Re: Search radius quality

larger radius helps find longer motion vectors - i.e. detect faster motion

Re: Search radius quality

I understand that but I am having some artifacts with "Average" and many with "Large" and I was hoping there is a way to have less artifacts on average or large searches by improving the detection method - is it possible to improve their 'quality' in a sense?

Re: Search radius quality

behold the black magic power of SVPflow (MVTools) limits and penalties big_smile

Re: Search radius quality

I have read this but I am having difficulties understanding it; are there lines that can be edited or changed to improve the mechanism that searches to limit for example the window blind artifacts?

Re: Search radius quality

the only ultimate solution for "window blind artifacts" is search radius = 1
however SVP's "adaptive" search radius makes it much better for larger values as well

Re: Search radius quality

By "adaptive" radius you mean options "Small" - "Average" - "Large" right? It does work well, but I was just hoping I can completely eliminate the artifacts that appear on "Average".

I was thinking something in terms of a more complex search method to avoid the blind artifacts that exist on "Average" but not on "Small" for example. There is nothing that can be done to improve the "Average" search - even at the expense of a lot of CPU resources?

Re: Search radius quality

by "adaptive" I mean negative values in SVPflow parameters
the actual search radius depends on the contrast value of the image block

Re: Search radius quality

I think you mentioned this the last time I asked something else regarding Search radius but it seems that I misunderstood you now and then. Can you please explain to me how I can change this parameter and try the "adaptive" setting you are mentioning?

Is it the "analyse.main.search.distance" on "override"? Or should I look for something different?

Re: Search radius quality

it IS already set to "adaptive"
"small" maps to "analyse.main.search.coarse.distance = -6", "average" = "-8", "large" = "-12"

Re: Search radius quality

I see, thanks. So there is nothing that can be done to improve the "Average" or "Large" setting regarding window blinds?

Re: Search radius quality

again - the black SVPflow magic, when each parameter affects resulting image in a very unclear way
it's like 10-way optimization problem when you never can say that "parameters set A is always better than parameters set B"

Re: Search radius quality

But is there a specific set of parameters which I can tamper with to maybe achieve a better result?

Also, while testing, I noticed the following:

There are 3 specific scenes on 3 different movies I run tests and all showed the same result. For some reason:

A) When Motion vectors grid is set to 12 px. Average 2 - Search radius has to be set to small to eliminate window blind artifacts.

B) When Motion vectors grid is set to 24 px. Large 2 - Search radius can be set to either Average or even Large and I get absolutely no window blind artifacts

However, even 16 px. Average 0 gives almost the same amount of window blind artifacts as 12 px. Average 2. I believe this is abnormal, how can there be no improvement on 16 px. Average 0 on either average or large but I get no window blind artifacts on 24 px. Large 2 on both average and large?

Would it not be more normal if for example:
0 window blind artifacts for Average should be coupled with 24 px. Large 2 and
0 window blind artifacts for Large should be coupled with 28 px. Large 2 or below?

Re: Search radius quality

can you attach a screenshot with blind artifacts?

Re: Search radius quality

I have uploaded screenshots from the 3 main movies I run window blind and other artifact tests. The movies are: The Dark Knight, Aliens, and The Matrix respectively. This does not mean that no other movies have similar artifacts, only that I use these as an example. All files are from their original blu-rays in .m2ts format, which means they are completely untouched quality wise.

On the right side of the screen, near the actor's face, over the lamp:
http://i.imgur.com/NsITu9B.jpg

On the right side of the screen, near the actor's face, under the lamp:
http://i.imgur.com/egRR8Xp.jpg

On the left side of the screen, the ventilation system:
http://i.imgur.com/0mF8Inu.jpg

On the left side of the screen, the first and second exhaust of the ventilation system:
http://i.imgur.com/gCDMQ2O.jpg

On the top left side of the actor's head and the top and lower top right side of the actor's head:
http://i.imgur.com/jCDqE2H.jpg

This relates to artifacts similar to the image above, on the top left side, but this also shows the merging that happens sometimes beyond window blinds towards other objects. For example the window blind artifact slightly merges with the actor's silhouette of white clothing creating a different kind of artifact:
http://i.imgur.com/b3HXMkQ.jpg

These screenshots are but a fraction of what is happening when the movies are playing because the artifacts come and go and appear in waves thus making them much more noticeable, especially on big screens. However, I did use "Large" Search radius in order to produce more artifacts so you can better study them. Using "Average" makes less artifacting but again on the same areas, while "Small" almost completely eliminates them (unnoticeable unless you are looking for them specifically).