1 Reply by max_pla3

(4 replies, posted in Using SVP)

Chainik wrote:

you may need to add "distributor()" at the end

Works perfectly! Many thanks! Just made a donation for this great software and support.

Go to forum: Using SVP

2 Reply by max_pla3

(4 replies, posted in Using SVP)

Thanks for the reply. The thing is that I see no difference between 2, 8 or even 16 threads. My CPU is an i7 860 with 4 Cores + HT. The GPU is a 7970.
In the task manager I see that the CPU is loaded to about 12% which corresponds to 1/8 of the total possible load. So this means that actually only on thread is shared across all cores. This is the case regardless of whether the gpu is enabled or not (gpu:0/1) and independent of the defined thread number in the script.

Now if I let the SVP Manager do the work, the load can increase up to 40%, denepnding on the clip, and the playback is smooth.

So the problem seems to be that my AviSynth script is not multithreaded despite SetMTMode(3,8). Any hint?

Thank you!

Go to forum: Using SVP

3 Topic by max_pla3

(4 replies, posted in Using SVP)

I have performance issues with the following AviSynth script:

SetMemoryMax(1024)

LoadPlugin("C:\Program Files (x86)\AviSynth\plugins\svpflow1.dll")
LoadPlugin("C:\Program Files (x86)\AviSynth\plugins\svpflow2.dll")

threads=2
SetMTMode(3,threads)
FFmpegSource2(FULL_PATH_TO_FILE)
SetMTMode(2)

orig_fps=FrameRate()

super_params="{gpu:1}"
super = SVSuper(super_params)

analyse_params="{}"
vectors = SVAnalyse(super, analyse_params)

smoothfps_params="{rate:{num:2,den:1}}"
SVSmoothFps(super, vectors, smoothfps_params, url="www.svp-team.com", mt=threads)

AssumeFPS(orig_fps)

It should slow down the video by a factor of two. The problem is that even a 720p video is not rendered with 60fps, if I comment out the last line.
BUT: Using the SVP Manager and MPC-HC, I can view the very same video with approx. 100 fps, if I increase the playback speed in the player.

What could be the issue?

In MPC-HC the EVR is already selected.

Go to forum: Using SVP

4 Topic by max_pla3

(3 replies, posted in Using SVP)

I'm experimenting with the algorithms and there's an issue with the interpolated frames. In the picture you can see three frames, whereby the first and the third frames are original and the second is interpolated with algorithm 1. The axe is moving up with a constant speed which means that the position of the axe in the second frame should be between those of frame one and three. But, the predicted position of the axe is much closer to the third frame (almost indentical) than it is to the first. This is more like frame doubling instead of interpolation.

Also, algorithm 2 renders exactly the same frame as algorithm 1. Though, if I increase threefold the frame number (num:3 instead of num:2), there is a difference between algorightms 1 and 2. But even then I can see that the axe in the interpolated frames is too close to the original frames 1 and 4 and there is a huge jump of the axe between the two interpolated frames 2 and 3.

Is there a way to fix it?

Here's my script:
**************************************************************************************
SetMemoryMax(1024)
LoadPlugin("C:\Program Files (x86)\AviSynth 2.5\plugins\svpflow1.dll")
LoadPlugin("C:\Program Files (x86)\AviSynth 2.5\plugins\svpflow2.dll")

threads=8
SetMTMode(3,6)
DirectShowSource(FILENAME).ConvertToYV12()
SetMTMode(2)

# All parameters set to defaults which means high quality frame doubling
super=SVSuper("{gpu:1}")
vectors=SVAnalyse(super, "{}")

SVSmoothFps(super, vectors, "{rate:{num:2, den:1, abs:false}, algo: 2}", url="www.svp-team.com", mt=threads)
# algo:
#       1 - sharp picture without any blending, moves pixels by motion vectors from next frame to current. Requires only backward motion vectors ("analyse.vectors: 2") so it's the fastest possible method.
#       2 - like 1st but moves pixels from the nearest (in terms of time) frame so it uses both backward and forward vectors. Recommended for 2D animations.
#      11 - time weighted blend of forward and backward partial motion compensations.
#      13 - same as 11th but with dynamic median added. Produces minimum artifacts but with noticeable halos around moving objects.
#      21 - 11th plus additional cover/uncover masking to minimize halos and improve frame edges.
#      23 - 21th plus extra vectors from adjacent frames for further decreasing of halos, can be less smooth than 21th.

#AssumeFPS(20)
#Levels(0, 1.3, 255, 0, 255)
**************************************************************************************

Go to forum: Using SVP