• Thanks: 6

Topic: Width of top coarse level - what is it? [Yes I see the description]

Width of top coarse level – for hierarchical search of motion vectors, at different levels different search options are used – particularly, at the last levels (the largest ones), more simple options are used to reduce the CPU load. The smaller this value is, the larger number of high levels will have the lower options.

Honestly, this description is incomprehensible. Can anybody outside of people working directly with the code make sense of this? There are so many adjectives it's impossible to decipher what this really means.

You can reread it 18 times but that last sentence is still impossible to make sense of.

The smaller this value is, the larger number of high levels will have the lower options.

Am I just stupid or is this a brain teaser??

Re: Width of top coarse level - what is it? [Yes I see the description]

let's take a 1920*1080 video as example

top (higher) level (#1) for the "hierarchical search" is the frame itself - 1920*1080
next one #2 - the frame reduced by 2 in each dimension - 960*540
#3 - 480*270
#4 - 240*135
bottom (lower) level #5 - 120*67

top level = "finest" level
all other bottom levels are "coarse" levels

the search parameters are _different_ for coarse and top levels
levels from #5 to #2 - more exhaustive search, level #1 - simplified search

with this "width of coarse level" option we can move this "line" down: levels 5 to 3 - exhaustive, levels 2 to 1 - simplified

and this is more important for 4K video

Re: Width of top coarse level - what is it? [Yes I see the description]

The description makes it really hard to understand. Does a smaller value make a 1080p video look better or worse?

  • Thanks: 1

Re: Width of top coarse level - what is it? [Yes I see the description]

Yeah, it doesn't seem like this was written by someone whose native language is English.  It seems to me as though Small will use the least CPU for this portion of the job, and Average will use a medium amount of CPU, and Large will use the most amount of CPU.  And I believe it is talking about how much CPU to use in the area where it matters the least, at the coarse end of the hierarchy.  Once it gets to the fine end of the hierarchy that is where spending more CPU will make more of a difference, or show better results.  I think what it means is that at this point of the job the amount of CPU that you spend will not be as important so you might not want to bother spending more CPU since it should affect the final result the least.  You can set it to Average or Large if you still want it to spend more CPU here, but the effect on the final result may be what is called diminishing returns.  You won't likely see a difference compared to the amount of CPU spent.

I could be wrong.