Search options

Hello,

Thanks for SVP4 which at home is more stable than SVP3.
However on my PC the function crooping results in a big drop in performance ... I find it weird :
In x4 and 1920x1080 (setting by default / 12 px. Average 2) with "cut off black bars" activated it is stable limit and completely unstable below 12 px. ... If I disable "cut off black bars" it is stable with 7 px. Small 2 !?
But this is not very serious since with my GTX650Ti I have no bug with the activated black bands (which I had with SVP3 and an HD6850).

I have some questions about SVP4.
In Auto mode there are 9 settings (fi_auto_adjust = -50/-40/-30/-20/-10/0/+10/+20/+30)

What are these settings in Manual Mode ?
Because regardless of the automatic settings selected when switching to manual mode, you have :
Motion vectors precision : Half pixel
Motion vectors grid : 12 px. Average 2
Decrease grid step : Disabled
Search radius : Average
Wide search : Average
Width of top coarse level : Large
Processing of scene changes : Repeat frame
Rendering device : Do not change

I suppose these settings do not change regardless of the automatic setting (fi_auto_adjust) ? :
Frames interpolation mode : Adaptive
Processing of scene changes : Repeat frame
Rendering device : Do not change

What are the manual settings for fi_auto_adjust = -50/-40/-30/-20/-10/0/+10/+20/+30 ?
Artifacts masking : ?
Motion vectors precision : ?
Motion vectors grid : ?
Decrease grid step : ?
Search radius : ?
Wide search : ?
Width of top coarse level : ?

Here is a real example (SVP X4) starting from 3 original frames of different algos, I also added SVP X2 1m + madVR Smooth Motion and Premiere CC 2017 Optical Flow.
My settings (smaller Motion Vector Grid for maximize artefacts) :

I also changed type=2 by type=4 (analyse.main.search.type)
2: Hexagon search (most lightweight)
3: Uneven Multi Hexagon (UMH)
4: Exhaustive search (most heavyweight)

SVP X4 Uniform-1m-1.5m-2m (Resized@50%)
SVP X4 Uniform-1m-1.5m-2m (Original Size)

Can you give us details on "Width of top coarse level" and "analysis.main.search.type" ?

juste for fun ...

Test on 60Hz screen with low motion blur (for example CRT or BenQ Z series) :
- You can compare a native 60fps@60Hz  video (1-2ms motion blur / No double Image effect  / Low Flicker / and smooth) and a video 30fps@60Hz (1-2ms motion blur / dramatic Double Image effect  / Low Flicker / and not smooth). This is the equivalent of blurbusters "Frame Rate Compare" tests.
- You can also compare with 30fps+1BFI@60Hz video (1-2ms motion blur / No double Image effect / Very High Flicker / and smooth). 30fps@60Hz vs 30fps+1BFI@60Hz is the equivalent of blurbusters "Black Frame Insertion" test.

Test on 60Hz screen with motion blur (most of LCD) :
- You can compare a native 60fps@60Hz  video (16.7ms motion blur / No double Image effect / No Flicker / and smooth) and a video 30fps@60Hz (33.3ms motion blur / No double Image effect / No Flicker / and smooth).
- You can also compare with 30fps+1BFI@60Hz video (16.7ms motion blur / No Double Image effect / High Flicker / and smooth).

Test on 120Hz screen with low motion blur (for example CRT or All LCD with BenQ blur reduction / LightBoost / ULMB) :
-You can compare a Native 120fps@120Hz video (1-2ms motion blur / No double Image effect / No Flicker / and smooth) to a video 60fps@120Hz (1-2ms motion blur / Visible double Image effect / No Flicker / and not completely smooth) and a video 30fps@120Hz (1-2ms motion blur / Dramatic quadruple Image effect / No Flicker / and not smooth). This is the equivalent of blurbusters "Frame Rate Compare" tests.
- You can also compare to the 60fps+1BFI@120Hz video (1-2ms motion blur / No double Image effect / Low Flicker / and smooth). 60fps@120Hz vs 60fps+1BFI@120Hz is the equivalent of blurbusters "Black Frame Insertion" test with two UFOs. 30fps@120Hz vs 30fps+3BFI@120Hz is the equivalent of blurbusters "Black Frame Insertion" test  with four UFOs.

Test on 120Hz screen with motion blur (LCD 120-144Hz With No Motion Blur Reduction) :
- You can compare a Native 120fps@120Hz video (8.3ms motion blur / No Double Image effect / No Flicker / and smooth) to a video 60fps@120Hz (16.7ms motion blur / No Double Image effect / No Flicker / and smooth) and a video 30fps@120Hz (33.3ms motion blur / No Double Image effect / No Flicker / and smooth).
- You can also compare the 60fps+1BFI@120Hz video (8.3ms motion blur / No Double Image effect / No Flicker / and smooth). And also compare the 30fps+3BFI@120Hz video (8.3ms motion blur / No Double Image effect / High Flicker / and smooth).

The 60fps video was obtained by accelerating (x2.5) of the original 24fps video.
The 30fps video was obtained by selecting one frame on 2 of the 60fps video.
The 120fps video was obtained by interpolating (SVP-x2) of the 60fps video.

Why cinema projectors inserts black frames ?

http://en.wikipedia.org/wiki/Movie_projector

Are there any means add the blackframes via GPU (shader) ?
Because avisynth script is too greedy ! (about 30% of my dual xeon system) ...

I confirm that the "testufo" are fully functioning !

Here extracted videos.

"compare frame rates:UFO" :
96 vs 48 vs 24fps @ 960pixels/s :
http://vandenk.free.fr/SVP/96-48-24fps(960).m2ts
120 vs 60 vs 30fps @ 960pixels/s :
http://vandenk.free.fr/SVP/120-60-30fps(960).m2ts
96 vs 48 vs 24fps @ 1920pixels/s :
http://vandenk.free.fr/SVP/96-48-24fps(1920).m2ts
120 vs 60 vs 30fps @ 1920pixels/s :
http://vandenk.free.fr/SVP/120-60-30fps(1920).m2ts

"Black Frame Insertion" (96 hz / 120 hz monitor) :
48fps(FrameDoubling) vs 48fps(+1BlackFrameInsertion) @ 1920pixels/s :
http://vandenk.free.fr/SVP/48doubvs48BFI.m2ts
60fps(FrameDoubling) vs 60fps(+1BlackFrameInsertion) @ 1920pixels/s :
http://vandenk.free.fr/SVP/60Doublvs60BFI.m2ts

note:
on previous videos, the written frames rates must be multiplied by 2 !
I made the capture @ 48fps and 60fps and doubled the frame rate when remux

Ok I download Hi-Definition Reference Disc 2008  Blu-ray 1080I AVC LPCM7.1 ... and I test !

No it's good test , if you print screen "Compare Frame Rates UFO" you see only one image for all framerate ...
I see this effect clearly in the movie @ 24fps (slow traveling for example) without FI

I think 24 -> 48fps (SVP FI) + black frame insertion is as smooth as 24 -> 96fps (SVP FI) with less artifacts and less CPU usage.

Chainik
it's hard to answer in general
there's a (complicate) function that transforms real frame time stamp

Ok, proceed by steps !
For "Double source frame rate (2x)" (24->48 , 25->50 , 30->60 , 48->96 , 50->100 , 60->120 ,...). This is good ? :

Edit - Recapitulation.
I added 1.5m algo and modify the presentation, the percentages match the distance with the previous Original Frame :

I also seeks to understand the options of SVP settings ...
I found some in these posts :
http://www.svp-team.com/wiki/Watching_anime
http://www.svp-team.com/forum/viewtopic.php?id=652
http://www.svp-team.com/forum/viewtopic.php?id=962
http://www.svp-team.com/wiki/Plugins:_SVPflow

Frames interpolation mode :

• A balance between smoothness strength and artifacts visibility. Controls the number of interpolated frames in resulting video.

OK for 24->60 conversion output will be :

is that correct ?

But for other conversion (x2,x3,x4,25->60,any refresh rate) what is the sequence ?