Avisynth Script for doubling video frame rates - Page 63 - AVS Forum
Forum Jump: 
Reply
 
Thread Tools
post #1861 of 1968 Old 01-01-2012, 03:44 PM
Newbie
 
gracemillian's Avatar
 
Join Date: Jan 2012
Posts: 1
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Hi guys, I've read the first 35 pages of this thread and have tried numerous scripts by widezu69, subjunk and have enjoyed suggestions by petitdragon and travolter and many others.

The point is, I have an Intel Core i7 920XM [mobile] and an ATI Raedon HD 5850m [mobile], could someone be kind enough to summarise what I've missed in the next 30+ pages?

PS: I've tried SVP, but I had issues with scrubbing [playing with the dot on the timeline, lol]
gracemillian is offline  
Sponsored Links
Advertisement
 
post #1862 of 1968 Old 01-02-2012, 03:44 AM
Member
 
Chainik's Avatar
 
Join Date: Apr 2011
Posts: 137
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 15
Quote:
Originally Posted by gracemillian View Post

I've tried SVP, but I had issues

SVP is just a GUI to control the very same avs-script. And all scripts here on last 20 pages are based on SVP libs
Chainik is offline  
post #1863 of 1968 Old 01-05-2012, 07:54 AM
Newbie
 
tigermaster's Avatar
 
Join Date: Jan 2012
Posts: 10
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Hi all, this is my first post, i'm following this thread since some months and i've read the first 15 pages plus the newest 6.

i have a core2duo E8600 @ 4Ghz + 4Gb ddr2 1066 + geforce GTX470 on Win7x64
Until yesterday i was using a couple of script for frame doubling that was working perfectly, using Avisynth 2.5.8 + mpc + coreavc + fddshow raw, all the stuff was updated to 6 months ago.

the scripts, for 720p, about 60% cpu

Code:
#mode 2d (multithread 720p)
SetMTMode(3,4)
ffdShow_source()
SetMTMode(3,2)
super=MSuper(pel=2, hpad=4, vpad=4)
backward_1=MAnalyse(super, chroma=false, isb=true, blksize=16, blksizev=16, searchparam=1, search=3, badrange=(-24))
forward_1=MAnalyse(super, chroma=false, isb=false, blksize=16, blksizev=16, searchparam=1, search=3, badrange=(-24))
backward_2 = MRecalculate(super, chroma=false, backward_1, blksize=8, blksizev=8, searchparam=1, search=3)
forward_2 = MRecalculate(super, chroma=false, forward_1, blksize=8, blksizev=8, searchparam=1, search=3)
MBlockFps(super, backward_2, forward_2, num=FramerateNumerator(last)*2, den=FramerateDenominator(last)*1, mode=2)
SetMTMode(1,2)
GetMTMode(false) > 0 ? distributor() : last
script for 1080p, up to 90% but only in very rare cases, for example on Trasformers imax version that uses full 1920*1080 resolution in some scenes, by the way i have no frame dropped, never (changed pel from 2 to 1, and pixel on recalculate from 8 to 16 compared to the above one).

Code:
#mode 2c (multithread 1080p)
SetMTMode(3,4)
ffdShow_source()
SetMTMode(3,2)
super=MSuper(pel=1, hpad=4, vpad=4)
backward_1=MAnalyse(super, chroma=false, isb=true, blksize=16, blksizev=16, searchparam=1, search=3, badrange=(-24))
forward_1=MAnalyse(super, chroma=false, isb=false, blksize=16, blksizev=16, searchparam=1, search=3, badrange=(-24))
backward_2 = MRecalculate(super, chroma=false, backward_1, blksize=16, blksizev=16, searchparam=1, search=3)
forward_2 = MRecalculate(super, chroma=false, forward_1, blksize=16, blksizev=16, searchparam=1, search=3)
MBlockFps(super, backward_2, forward_2, num=FramerateNumerator(last)*2, den=FramerateDenominator(last)*1, mode=2)
SetMTMode(1,2)
GetMTMode(false) > 0 ? distributor() : last
i have made custom resolutions for my monitor so i don't use reclock at all with the above scripts, and i only double frames as you can see from the scripts, so basically i run the stuff at 48fps.

Yesterday i've come back to this thread and read some of the latest news, i've seen avysynth 2.6, also i've tried installing SVP but it has too much stuff and i would like to manage everything by myself until there are some real advantages in terms of quality or performances using SVP and the other renderer it has installed.

So i've updated avisynth and all the stuff in the "Files_8-8-11.zip" and i've tried to use the script provided by widezu69 some pages ago, main problem is that my pc doesn't have enough power to manage that script, i had to modify some values, and still i'm very high on cpu and sometimes i drop some frames:

svp_flow_lib="C:\\Program Files (x86)\\Avisynth 2.5\\plugins\\libflowgpu.dll"
svp_scheduler=true
SetMTMode(1,15)
V = ffdshow_source()
SetMTMode(2)
S = V.MSuper(pel=1, hpad=16, vpad=16, rfilter=4)
A = S.MAnalyse(isb=true, blksize=16, overlap=4, plevel=0, pelsearch=3, search=3, searchparam=1, lambda=2000, lsad=1600)
B = S.MAnalyse(isb=false, blksize=16, overlap=4, plevel=0, pelsearch=3, search=3, searchparam=1, lambda=2000, lsad=1600)
C = S.MRecalculate(A, blksize=16, overlap=2, search=3, searchparam=1)
D = S.MRecalculate(B, blksize=16, overlap=2, search=3, searchparam=1)
V.MSmoothFps(S, C, D,num=48, den=1, algo=13, block=false, sadgamma=2, sadml=0, ml=0, blend=true)
GetMTMode(false) > 0 ? distributor() : last

i have a couple of question, i've tried to modify num from 60 to FramerateNumerator(last)*2 but it's not recognized by avisynth, so i've changed it to 48, by the way i had to install reclock cause my audio was going out of sync, don't know if the sync issue is caused by high cpu usage, i need a more relaxed script to test it out, so in the new command V-MsmoothFps we can't use the above string?
About reclock i've putted it on "external filters" on mpc, at the bottom, is that right ? the chain is coreavc, then ffdshow raw then reclock.

EDIT: Reclock must be putted before ffdshow raw in order to let it sync the audio, just tested myself.

Also that SetMTMode(1,15) is referred to the GPU right? but then when it uses mvtools commands it goes back to the cpu, so SetMTMode(2) is ok or i should write (2,2) or (2,4) ?
Lastly, is there a more relaxed script or some more tweaks i can do on that one in order to reduce cpu usage a bit more? or maybe am i missing something? my gpu uses 10% power, any chance to let it use the gpu a bit more and a bit less cpu?
As an alternative is there an even better scripts that removes even more artifacts compared to the 720p (i mean using Mblock) one i currently use? i still have some cpu free on that script

EDIT: is there a way to use mflowfps or msmoothfps without using libflowgpu? i would like to try it using cpu, but the old command MVFlowFps is not recognized anymore, has it been changed in the avs 2.6 ?

thanks for this thread it really made my days in the last months.
tigermaster is offline  
post #1864 of 1968 Old 01-06-2012, 07:03 AM
Newbie
 
mustamyy's Avatar
 
Join Date: Sep 2009
Posts: 1
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Quote:
Originally Posted by tigermaster View Post

Lastly, is there a more relaxed script or some more tweaks i can do on that one in order to reduce cpu usage a bit more? or maybe am i missing something? my gpu uses 10% power, any chance to let it use the gpu a bit more and a bit less cpu?

EDIT: is there a way to use mflowfps or msmoothfps without using libflowgpu? i would like to try it using cpu, but the old command MVFlowFps is not recognized anymore, has it been changed in the avs 2.6 ?

thanks for this thread it really made my days in the last months.

You can use libflowsse for cpu only. First you want to use more gpu and then you don't, so which one is it? I don't see any reason not to use libflowgpu (well ok, my laptop overheats and crashes with it)
You don't have to use SVP manager, you can still tweak everything by hand.
Just be sure to use also SVP version of mvtools2.dll
mustamyy is offline  
post #1865 of 1968 Old 01-07-2012, 06:49 PM
Newbie
 
tigermaster's Avatar
 
Join Date: Jan 2012
Posts: 10
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Quote:
Originally Posted by mustamyy View Post

You can use libflowsse for cpu only. First you want to use more gpu and then you don't, so which one is it? I don't see any reason not to use libflowgpu (well ok, my laptop overheats and crashes with it)
You don't have to use SVP manager, you can still tweak everything by hand.
Just be sure to use also SVP version of mvtools2.dll

just for testing, i remember i've already made some tests with and without libflowgpu using another script and i saw no significant changes in terms of cpu load, but as it's strange i would like to repeat the test with this script.
about mvtools2.dll the one included in widezu69 zip file is 2.5.11.6, while on my hd i have 2.5.11.7. better if i keep widezu69 suggested one or i can use the newer one?
there are no info on mvtools2 website about this svp versions..
tigermaster is offline  
post #1866 of 1968 Old 01-14-2012, 08:12 AM
Newbie
 
tigermaster's Avatar
 
Join Date: Jan 2012
Posts: 10
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Ok i've done some more tests, as i said earlier it looks that libflowgpu uses more cpu than libflowsse at least on my pc, the same script uses from 20 to 25% less cpu when i use libflowsse, i don't know why, probably the processing power required to send and receive instructions from the gpu are higher than let it do everything from the cpu itself.

my chain is coreavc+reclock+ffdshow raw using MadVR as renderer, all filters are on "external filters", including reclock
About MadVR i've noticed it uses same cpu as EVR custom but i had near 0 frame dropped even with 95% cpu in use while with EVR custom i have much more frame dropped..

this is the script i've used for testing:

#mode 4b (GPU new)
svp_flow_lib="C:\\Program Files (x86)\\Avisynth 2.5\\plugins\\libflowgpu.dll"
svp_scheduler=true
SetMTMode(1,15)
V = ffdshow_source()
SetMTMode(2)
S = V.MSuper(pel=1, hpad=16, vpad=16, rfilter=4)
A = S.MAnalyse(isb=true, blksize=16, overlap=4, plevel=0, pelsearch=3, search=3, searchparam=1, lambda=2000, lsad=1600)
B = S.MAnalyse(isb=false, blksize=16, overlap=4, plevel=0, pelsearch=3, search=3, searchparam=1, lambda=2000, lsad=1600)
C = S.MRecalculate(A, blksize=16, overlap=2, search=3, searchparam=1)
D = S.MRecalculate(B, blksize=16, overlap=2, search=3, searchparam=1)
V.MSmoothFps(S, C, D,num=48, den=1, algo=13, block=false, sadgamma=2, sadml=0, ml=0, blend=true)
GetMTMode(false) > 0 ? distributor() : last

with a 720p source with gpu i'm at 90/95% cpu while with sse i'm at 70/75%, in fact with sse i've been able to set the block size back to 8 for recalculate, and still it's using less cpu than the other one.

this is the sse version:

#mode 5a (SSE 720p)
svp_flow_lib="C:\\Program Files (x86)\\Avisynth 2.5\\plugins\\libflowsse.dll"
svp_scheduler=true
SetMTMode(1,4)
V = ffdshow_source()
SetMTMode(2)
S = V.MSuper(pel=1, hpad=16, vpad=16, rfilter=4)
A = S.MAnalyse(isb=true, blksize=16, overlap=4, plevel=0, pelsearch=3, search=3, searchparam=1, lambda=2000, lsad=1600)
B = S.MAnalyse(isb=false, blksize=16, overlap=4, plevel=0, pelsearch=3, search=3, searchparam=1, lambda=2000, lsad=1600)
C = S.MRecalculate(A, blksize=8, overlap=2, search=3, searchparam=1)
D = S.MRecalculate(B, blksize=8, overlap=2, search=3, searchparam=1)
V.MSmoothFps(S, C, D,num=48, den=1, algo=13, block=false, sadgamma=2, sadml=0, ml=0, blend=true)
GetMTMode(false) > 0 ? distributor() : last

I've tested also a script with mflow, it give really nice results aswell, with minimal artifacts, i prefer msmooth but results are very similar, i had to remove the third pair as it was too much for my cpu, this one uses the same cpu as the above one with a 720p input:

#mode3b (mflow low)
SetMTMode(1,4)
V = ffdshow_source()
SetMTMode(2)
S = V.MSuper(pel=2, hpad=0, vpad=0, rfilter=4)
A = S.MAnalyse(isb=true, blksize=16, plevel=0, search=3, searchparam=1, badrange=(-24))
B = S.MAnalyse(isb=false, blksize=16, plevel=0, search=3, searchparam=1, badrange=(-24))
C = S.MRecalculate(A, blksize=8, search=3, searchparam=0, thSAD=200)
D = S.MRecalculate(B, blksize=8, search=3, searchparam=0, thSAD=200)
V.MFlowFps(S, C, D, num=FramerateNumerator(V)*2, den=FramerateDenominator(V)*1, mask=2, ml=10000)
SetMTMode(1)
GetMTMode(false) > 0 ? distributor() : last

I'll continue reading (page 25 now) the entire thread but i can say i'm already really happy with this solution, i think i will abandon Mblock, as results with this commands are much better, even with pel=1 and with only 2 pairs.
Now i would like to find something for a 1080p, also because i want to try frame doubling for squashed SBS 3D movies.

I have some questions:

-Unfortunately i was not able to set pel=2 with sse, it crashes mpc, does anyone knows why?

-Does anyone have done tests with 3d stuff?

-Where i can find the scripts used by SVP ?

-Why num=FramerateNumerator(last)*2, den=FramerateDenominator(last)*1 doesn't works on MSmoothFps ? How can i tell avisynth to double frames on that command without inserting a fixed integer ?

thanks

my detailed cfg just for information:

Code:
Core2Duo E8600 @ 4.06Ghz
4Gb DDR2 @ 1080Mhz
Geforce GTX470
Samsung display 2233rz 1680*1050 120Hz

software versions:
Win7x64
Nvidia driver 280.26
libflowsse 1.3.0.0/3.0.2
libflowgpu 1.2.6.0/3.0.5.1
mvtools2 2.5.11.8
Avisynth 2.6.0.2
MPC-HC 1.4.2677.0
ffdshow rev3828 ICL10
reclock 1877
MadVR 0.8
CoreAvc 3.1 build 8441
Haali media splitter 1.11.288.0
and my crappy svpmark benchmark
my DPC latency is around 100 microseconds

Code:
Test summary
-----------------------
  Date: 2012-01-05T22:24:15
  CPU:  Intel Core2 Duo E8600 @4500 MHz [2 threads]
  GPU:  NVIDIA GeForce GTX 470
  Mode: HD + GPU [5 threads]

Overall scores
-----------------------
  Synthetic CPU:                  MC633
  Synthetic GPU:                  MG920
  Real-life:                      HG1394

Details: synthetic
-----------------------
  CPU: compose (single-threaded): 521
  CPU: compose (multi-threaded):  1016
  CPU: search (single-threaded):  234
  CPU: search (multi-threaded):   460
  GPU: system -> GPU transfer:    733
  GPU: GPU -> system transfer:    693
  GPU: calculations:               652
  GPU: total score:               2838

Details: real-life /HD
-----------------------
  decode video:                   11.39x (273.3 fps)
  48 fps - vectors search:        0.72x (34.5 fps)
  60 fps - frame composition:     1.92x (115.2 fps)
  48 fps - [SVP] fastest:         3.40x (163.3 fps)
  48 fps - [SVP] simple 1:        2.20x (105.6 fps)
  60 fps - [SVP] good:            1.15x (68.9 fps)
  60 fps - [SVP] high:            0.94x (56.6 fps)
  60 fps - [SVP] highest:         0.49x (29.2 fps)
  72 fps - [SVP] simple 2:        1.87x (134.6 fps)
tigermaster is offline  
post #1867 of 1968 Old 01-15-2012, 05:28 AM
Member
 
Chainik's Avatar
 
Join Date: Apr 2011
Posts: 137
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 15
tigermaster
Quote:
as i said earlier it looks that libflowgpu uses more cpu than libflowsse at least on my pc
....
and my crappy svpmark benchmark

Something's wrong with you setup - your results are much lower than Core2 Duo E8400 @3600 + GeForce 9800 GTX. Take a look at "good" preset fps: 83.6 vs 68.9.
Try to rollback to 275 NV driver.
Chainik is offline  
post #1868 of 1968 Old 01-17-2012, 11:57 AM
Newbie
 
tigermaster's Avatar
 
Join Date: Jan 2012
Posts: 10
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Quote:
Originally Posted by Chainik View Post

tigermaster

Something's wrong with you setup - your results are much lower than Core2 Duo E8400 @3600 + GeForce 9800 GTX. Take a look at "good" preset fps: 83.6 vs 68.9.
Try to rollback to 275 NV driver.

Thanks a lot for your suggestion, i've installed 275.33 and now is ok, it's better with libflowgpu, got 98,3 at "good" preset, i'm very near to be able to run frame doubler for some 1080p contents, cropping only 120pixel each side and 16pixel on top/bottom, as i have a 1680*1050 monitor.
quite good considering i have only a dual core cpu.

My gpu now uses up to 30% while before i have never seen that over 10%.

Continue testing..
tigermaster is offline  
post #1869 of 1968 Old 01-21-2012, 05:43 AM
Member
 
dzid_'s Avatar
 
Join Date: May 2010
Location: Poland
Posts: 32
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Quote:
Originally Posted by khordo View Post

Hy DZID, i tried your GUI on my Win7X64 with no success
The conversion is running and suddenly the system stop, No process running at all...You need to click on abort and restart all the process. But next time the same issue
Do you have any idea ?

It happens to me when I am low on memory (ram). If you don't have much ram try to add Setmemorymax(700) to the script. (or even less than 700)
dzid_ is offline  
post #1870 of 1968 Old 02-05-2012, 12:06 PM
AVS Special Member
 
sanderdvd's Avatar
 
Join Date: Nov 2001
Location: The Netherlands, Europe
Posts: 2,196
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 54 Post(s)
Liked: 24
Hi guys,

I own a JVC RS25 projector which has very bad motion handling. Want to give frame doubling a try. I m running:

W7 x64
MPC-HC
LAV filters with CUVID hardware decoding enabled
madVR 0.80 as video output renderer
NVIDIA GTX460 oc-ed with lastest driver
display device: JVC RS25 projector

Content I m running is 1080p Blu-Ray material @23,976.

Could someone help me on my way?
sanderdvd is offline  
post #1871 of 1968 Old 02-06-2012, 03:25 AM
Newbie
 
ceth's Avatar
 
Join Date: Jan 2010
Posts: 14
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Use SVP (smooth video project). Easiest way to try it, install and you are set
ceth is offline  
post #1872 of 1968 Old 02-09-2012, 01:10 AM
AVS Special Member
 
sanderdvd's Avatar
 
Join Date: Nov 2001
Location: The Netherlands, Europe
Posts: 2,196
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 54 Post(s)
Liked: 24
I gave SVP try because I hate the motion stutter on my JVC RS-25 projector. I installed everything I need and it works almost perfect exept for madVR in combination with SVP:

My projector supports 1080P23, 1080P24, 1080P50 and 1080P60. I ve set my desktop resolution in my NVIDIA control panel to 1920x1080@60 and SVP to frame rate to screens refresh rate. The problem is that I get a lot of dropped frames in madVR (one dropped frame every 2 minutes reported my madVR). I can understand this (I think) because madVR needs an exact match of the fps of the content that is playing (or a double of that I think). So without SVP i set my NVIDIA control panel to EXACTLY 1920x1080@23,976 (when running a Blu-Ray movie) and I get 1-4 dropped frames during a 2 hour movie which is great. When using SVP it of course has no sense to use it without setting my NVIDIA output to 60Hz. Will I not be able to use madVR and SVP?
sanderdvd is offline  
post #1873 of 1968 Old 02-09-2012, 02:24 AM
Member
 
dzid_'s Avatar
 
Join Date: May 2010
Location: Poland
Posts: 32
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
I would try Reclock.
I don't know your setup but with high cpu usage produced by SVP, madVR tends to drop frames anyway.
You can also try madVR with some lower scaling values - it might be cpu problem to upscale image at 60fps.
You can also try some other renderers with up-scaling capabilities. In my opinion, if you don't bother with artefacts produced by SVP you shouldn't care much about slightly worse, but more reliable, renderers. Currently I am using custom VMR9 Renderless (in potplayer) with bicubic scaler settings.

In other hand to obtain better 60fps interpolation you should play with Widezu's scripts or if you want the best possible quality and no frame drops, take a look at SubJunk's Interframe, which performs not real time conversion.
dzid_ is offline  
post #1874 of 1968 Old 02-09-2012, 10:56 AM
Senior Member
 
johnifehr's Avatar
 
Join Date: May 2008
Posts: 461
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
What are your computer specs?
johnifehr is offline  
post #1875 of 1968 Old 02-12-2012, 03:06 AM
Newbie
 
tigermaster's Avatar
 
Join Date: Jan 2012
Posts: 10
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Quote:
Originally Posted by sanderdvd View Post

(one dropped frame every 2 minutes reported my madVR)

Where do i have to look to see how many frames madvr is dropping? on MPC-HC when i set madvr as renderer statistics always says 0 dropped frames, wich is not true obviously.

I'm testing SVP and i have to admit is very well done... nice job!
tigermaster is offline  
post #1876 of 1968 Old 02-14-2012, 08:58 AM
AVS Special Member
 
sanderdvd's Avatar
 
Join Date: Nov 2001
Location: The Netherlands, Europe
Posts: 2,196
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 54 Post(s)
Liked: 24
Quote:
Originally Posted by dzid_ View Post

I would try Reclock.
I don't know your setup but with high cpu usage produced by SVP, madVR tends to drop frames anyway.
You can also try madVR with some lower scaling values - it might be cpu problem to upscale image at 60fps.
You can also try some other renderers with up-scaling capabilities. In my opinion, if you don't bother with artefacts produced by SVP you shouldn't care much about slightly worse, but more reliable, renderers. Currently I am using custom VMR9 Renderless (in potplayer) with bicubic scaler settings.

In other hand to obtain better 60fps interpolation you should play with Widezu's scripts or if you want the best possible quality and no frame drops, take a look at SubJunk's Interframe, which performs not real time conversion.

thanks for your reply.
I can t use ReClock (read, has no benefit) because I need bitstreaming for my D-Box motion decoder that I use.

With the help of MAG79 and Chainik I now have SVP working 100% correctly with 3! dropped frames max during a 2 hour movie.

Only problem I have is lipsync but I ll figure that out later (could be NVIDIA driver+SVP combination related).

Once I get the lip sync problems figured out the next step will be the actual configuration of SVP. Last night I tried some first changes in the original settings and I noticed that almost every setting has it s affect on the visible artifacts. Is there a list somewhere where I can see a full explanation of every single setting that can be changed in the expert mode of SVP.
sanderdvd is offline  
post #1877 of 1968 Old 02-15-2012, 11:44 AM
AVS Special Member
 
sanderdvd's Avatar
 
Join Date: Nov 2001
Location: The Netherlands, Europe
Posts: 2,196
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 54 Post(s)
Liked: 24
If someone could help with some extensive explanation of the next settings and questions about it I would really appriciate that! I know it will take some time to answer them all but I would be thankful eternally

1. Smoothness Improvement
- Normal (an extra option then with the smoothness improvement option 'By Blocks' comes available which is called Blocks Overlapping).
- By Blocks
- Advanced (different SVP-shaders to choose from, Contour Artifacts Suppression becomes available).

Questions:
Is By Blocks the leasts accurate smoothness option which causes the most artifact and is the least CPU intensive?

Which of the three offers the best picture quality and gives the least artifacts if configured correctly and CPU is fast enough?

With the smoothness normal option, what are the differences between the selectable SVP-shaders Standard, Simple and Simple Bidirectional. And what are the artifacts of each of those three that can become visible?

With the smoothness advanded option, what are the differences between the selectable SVP-shaders Original and Complex. And what are the artifacts of each of those three that can become visible?

2. Framerate Change:
What can be told as explanation for all the different types of framerate changes you can select here besides of course differences in CPU usage? What do they exactly do? Which gives the least artifacts with best smoothness? What is 1m and 2m? etc.

3. Interpolated Frames Size
What does it do exactly and is Full size the best and maybe only option you want to pick here?

4. Search Radius
Is 4 pixels to most accurate (and of course most CPU intensive) with the least artifacts you can get? What does it do exactly?

5. Search Types
How do the 4 selectable options here (Logarithmic, Hexagon, Uneven Multi Hexagon and Exhaustive) different from each other and which gives the leasts artifacts, best smoothness etc?

6. Motion Estimation Accuracy
It seems that that half or quater-pixel is very CPU intensive. What does it do exactly and what about the artifacts of them?

7. Recalculation Of Vectors
What happens with the picture when you select to recalculate with blocks dividing in this setting? How does this affect smoothness and artifacts?

8. Blocks Size
Does 32x32 gives the best smoothness and least artifacts or 8x8? What does it do exactly?

9. Blocks Overlapping
What does it do exactly and do you get less or more artifacts when choosing this option?

10. Contour Artifacts Suppression
Do you get the least artifacts when choosing high here? Are there other things to tell about this setting?

11. Bad Areas Artifacts Suppression
Do you get the least artifacts when choosing high here? Are there other things to tell about this setting?

12. Bad Areas Artifacts Mask
What do the options Light, Normal and Sharp do and which gives the least artifacts?

13. Processing Of Scene Change
What is the difference between Frame Blending and Frame Repeating?

14. Reduce Resolution
Was this option created for very low CPUs?
sanderdvd is offline  
post #1878 of 1968 Old 02-15-2012, 07:29 PM
AVS Special Member
 
greg1292's Avatar
 
Join Date: Oct 2004
Location: FORT WAYNE,INDIANA
Posts: 1,062
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 22 Post(s)
Liked: 18
Sanderdvd:

I don't use madVr but MPHC tester build from Doom9 x86 AVX: http://www.mediafire.com/?ceodtrhjc20qb1d which my cpu supports.
Use the SSEE2 build for other processors.
revision 4041; and here are some of the settings I use with the best results
for me so far.

EVR CUSTOM PRESET
MITCHELL-NETRAVALI CUBIC6 set to optimal renderer settings
with everything in the internal filters unticked.

External filters are:
lav splitter
file source(Async)
lav audio decoder
lav video decoder dxva2(copy-back) I use ATI
ffdshow raw all set to prefer
Halli media splitter set to block


SVP 3.06
1920X1080@24
advanced
23.complex
24-60 2m (1.5)
full size
2 pixels
hexagon
to half pixel
do not recaculate
16x16
.25 block
normal
below normal
normal
frame blending
do not reduce

For me the revision 4041 AVX build by JanWilliams32(Huge Props) gives me everything I am looking for in a media player and the 32 point floating surfaces and 256 point look up and dithering capabilities. There are alot of
great choices but am very happy with Amd FX-4100 and ATI 6570.

Tried madVr but alway's go back EVR CUSTOM.

Good luck and hope this helps

 Specialize in maximizing performance on a beer budget!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

greg1292 is online now  
post #1879 of 1968 Old 02-16-2012, 11:49 AM
AVS Special Member
 
sanderdvd's Avatar
 
Join Date: Nov 2001
Location: The Netherlands, Europe
Posts: 2,196
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 54 Post(s)
Liked: 24
thanks for your reply!

What I m really looking for is explanation to the questions I asked. I want to know WHAT exactly all the specific settings do. (how they affact artifacts etc.)
sanderdvd is offline  
post #1880 of 1968 Old 02-17-2012, 06:21 AM
Newbie
 
tigermaster's Avatar
 
Join Date: Jan 2012
Posts: 10
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
sanderdvd can you please tell me where to look to see frame dropped using madvr?
really can't find it.
thank you

About specific settings explaniation i have only partial answers, everything imho and please someone correct me if i'm saying something wrong.

There is a bar on settings, so that means the higher the bar, the higher cpu needed obviously.

1-i know only "by block" here, is the older script used at first (mblockfps), it's faster but gives more artifacts usually, avoid this if you can.

2-framerate change. if you are able to create custom resolution for your display (47,952 and 50hz are the most common) the "double framerate option" is the best choose, it gives you enough smoothing and you will save some cpu, so you can increase other parameters.

3-it changes the resolution of interpolated frame size, it doesn't change much in terms of processing power, i've tried it on a 1080p and results are good also with 50% reduction, avoid this for lower resolution movies.

6-it's very hard to explain motion estimation with few words, is the method used by codecs to save data calculating the next frame using motion vectors, have a look to qpel and motion compensation on wiki. Quarter pixel gives the best calculation.

7-Same as above it refers to re-calculating vectors for motion comensation, if i'm not wrong i think this is not needed if you use only "double frame rate (2x)"

8-the smaller it is the least artifacts. By the way on 1080p you can try keeping it a bit higher, while on slower resolutions it's best to use 8x8, obviously if you can keep 8x8 always.

11-it blurs faster interpolated frames in order to reduce artifacts, the higher it is, the more it blurs, i suggest to keep it to below normal, or low, because i've noticed that with higher values it blurs also things that should not be blurred.

13- with frame blending it interpolates also scene changes, while with frame repeating it tries to understand when there is a scene change disabling interpolation.

14- you can try go to 720p if you have a 1080p.

Finally the only thing i can say is test by yourself, try change some parameters and see a part of a movie where there are a lot of scrollings o fast scenes, and see results, you can watch also frame by frame with MPC-HC just pause and then ctrl + arrows let's you able to go frame by frame, make some screenshot if you want to compare and choose what''s best for you, or for you current movie.

if someone else can help us with more detailed informations it's more than welcome!!

If you haven't done it already, install coreavc or lavcuvid in order to save some more cpu and use ffdshow as raw.

If you are still not happy with realtime smoothing you can always use Subjunk's Interframe and do it offline, there's a nice guide here, else you can try Interframe GUI.
Interframe's default options gives really nice results, if you are unhappy with the smoothing and you don't care about some rare artifacts you can try the tuning "smooth" option.
tigermaster is offline  
post #1881 of 1968 Old 02-17-2012, 08:22 AM
Newbie
 
tigermaster's Avatar
 
Join Date: Jan 2012
Posts: 10
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
I've tried to do some interpolated frames manually between 2 frames taken froma movie, using Sqirlz Morph, is a morphing tool that is used to morph 2 images togheter, usually to make face morphing.
I've manually added some warp points on some edges, results are unbelievable, i'm wondering if maybe some day someone will be able to automate this process, it should add warp point automatically based on edges, interpolated frames will be nearly perfect in this way.
Look at gifs in the zip, the first one is a standard morph, is something similar to what avisynth does actually, while the second one is a warp+morph, amazing!!
maybe one day...
It's a simulation with 100 frames between 2 real frame from the movie, a bit extreme, but is just to see how it works..

http://www.mediafire.com/?srubg2ykbd8aj1t
tigermaster is offline  
post #1882 of 1968 Old 02-17-2012, 12:35 PM
AVS Special Member
 
sanderdvd's Avatar
 
Join Date: Nov 2001
Location: The Netherlands, Europe
Posts: 2,196
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 54 Post(s)
Liked: 24
thxz for your reply tigermaster!
Pressing CTRL+J shows the dropped frames and presentation glitches.

2.
My display device (JVC RS25 projector) does not support 47,952 and 50, only 23.976, 25.000, 59.000 and 60.000 and everything between 59.000 and 60.000. On the SVP forum they suggest 1M of 2M but I don t know what that does and did not get an explanation on the forum about this yet. I m now running it on 'to screen refresh rate'. This results in very much smoothness but the downsides are more artifacts and higher CPU usage.

3.
So for this question full gives the best PQ?

6.
Noticed that quarter pixel is very CPU intensive but I might win something when changing 'framerate change' to something else then 'so screen refresh rate'.

8.
I only play 1080p content, what will be the actual difference when choosing the 'safe' (less artifacts) setting 8x8 or choose 16x16?

11.
So low is ok to use here.

13.
Why would someone not want interpolation during scene changes?

14.
? this will downscale 1080p content to eg. 720p? That s not what I want right?

Already using LAV with CUVID enabled which indeed saves my a lot of CPU. Tried subjunks option but really like the 'comfort' of real-time interpolation SVP does.
sanderdvd is offline  
post #1883 of 1968 Old 02-22-2012, 04:57 PM
Senior Member
 
widezu69's Avatar
 
Join Date: Feb 2010
Posts: 319
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Ladies and Gentlemen, hello, it has been a long time. Too long.

Let's jump straight into it shall we? First of all, for those using SVP, I suggest using the GPU accelerated option, if you can, for the best performance.

The best IMHO options for SVP:
Smoothness Improvement: Normal, the performance trade offs with Advanced make it not worth it.

SVP Shader: 13. Standard

Framerate: This is a tricky one. Usually, I would be inclined to go with refresh rate. It depends on what frame rate your video is and what refresh rate your monitor/projector can do. If you are playing a 23.976/24fps movie, I would set your monitor to 59.97/60Hz and use the refresh rate option. If your movie is 25fps, then set your monitor to 50Hz. As long as your refresh rate is a multiple of the original frame rate of your video, you will save lots of CPU because the algorithm isn't performing complex calculations to achieve the video frame rate. For example, a 25fps video @60hz, the algorithm must multiply by 12 then divide by 5. 24fps is a different case, despite no dividing into 60, it works by multiplying by 5 = 120 then just halving it. I only watch 24fps movies therefore my refresh rate is 60Hz and my SVP setting is "To screen refresh rate"

Interpolated frames size: Full

Search radius: 2 pixels. You can go down to 1 if flickering fence and window patterns are distracting but at a loss of smoothness. Any higher is smoother but you get lots more artifacts.

Search type: Exhaustive

Motion accuracy: Half-pixel. Quarter is was too stressful on the CPU

Recalculate: with block dividing

Blocks size: 16x16 if on 720p, 32x16 if on 1080p, 8x8 on 480p and anything smaller.

Blocks overlapping: 0.25 of block

Artifact suppression: Nothing, any increase will hurt smoothness, plus shader 13 does a decent job anyways.

Artifact mask: Sharp

Processing of scene change: Frame blending. This controls what the algorithm does when there is a change of camera angle/scene/shot however you want to call it. This usually works quite well when set to repeating which means that it will not try and make an interpolated frame between two completely different shots. However, sometimes, it will regard fast moving sequences as scene changes and refuse to interpolate anything when it should. Leave it at blending is the best.

Reduce resolution: Do not reduce.

Don't bother checking blending if you are using the refresh rate function anyways.
widezu69 is offline  
post #1884 of 1968 Old 02-23-2012, 03:48 AM
Member
 
Chainik's Avatar
 
Join Date: Apr 2011
Posts: 137
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 15
Quote:
Originally Posted by widezu69 View Post

Reduce resolution: Do not reduce.

... unless you watch 1080p video on a smaller screen. In this case decreasing video frame size to actual screen size before motion interpolation can give a huge performance boost.
Chainik is offline  
post #1885 of 1968 Old 02-23-2012, 09:00 AM
Newbie
 
tigermaster's Avatar
 
Join Date: Jan 2012
Posts: 10
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Quote:
Originally Posted by widezu69 View Post

Recalculate: with block dividing

Welcome back, and thanks for all your replies!

I'll check your suggestions for next movie.
Just a question about motion vector recalculation, if i use "double frame rate" is vector re-calculation still needed? I mean, we have already vector informations for both the frame before and the frame after the one that has to be calculated by SVP, so what's the point of recalculating? or am i missing something ?

Also about the block sizes, i've noticed that with 32*16 during scrollings the halo around static object is bigger than if i use 8*8, even on a 1080p source, are your suggestion based on saving cpu? it looks i get better results if i keep 8*8 all the time (i really can't keep 8*8 on a 1080p with my cpu btw).

thanks
tigermaster is offline  
post #1886 of 1968 Old 02-23-2012, 04:01 PM
Senior Member
 
widezu69's Avatar
 
Join Date: Feb 2010
Posts: 319
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
@tigermaster,

In answer to your first question, the vector recalculation has nothing to do with how many frames you wish to output. Recalculation is basically a second pass. The algorithm scans the changes in the frame, then performs the scan again but also using the vectors from the first pass thus improving frame accuracy. Think of when doing x264 encoding, 2 passes is always better than one even if it takes more time.

The reason why I chose that particular block size is for 1080p is due to the size of the block. The configuration above generally works best for 720p stuff. I'm using 16x16, which when recalculated is using 8x8 in the second pass (that's how it works) overlapping using 0.25 blocks so 4x4 and 2x2 respectively. When using 8x8, that gets reduced to 4x4 the second pass when overlapping, that because 2x2 and 1x1. Unfortunately, the overlap function in MVTools2 does not like 1x1 very much hence I don't recommend using a 8x8 configuration.

720p video uses 16x16 so it makes sense for 1080p, which is 1.5, times larger to use 24x24 but that is not supported so, the next best thing is 32x16 which is the same in terms of surface area.

I assume you are using MPC-HC using ffdshow. Might I suggest that you disable SVP and exit from it completely and try using a script just to see whether it makes any difference.

Basically, play a video, pause, right click and go to Filters, ffdshow video filter, properties and go to the Avisynth tab. Make sure Avisynth, YV12 and Buffer back/ahead are all checked and nothing else. Buffer back/ahead should be 0/24.

Put this script in the area below it:

svp_flow_lib="C:\\Program Files (x86)\\SVP\\Plugins\\MVTools2\\libflowgpu.dll"
svp_scheduler=true
SetMTMode(1,15)
V = ffdshow_source()
SetMTMode(2)
S = V.MSuper(pel=2, hpad=16, vpad=16, rfilter=4)
A = S.MAnalyse(isb=true, blksize=16, overlap=4, plevel=0, pelsearch=3, searchparam=2, search=3, lambda=2000, lsad=1600)
B = S.MAnalyse(isb=false, blksize=16, overlap=4, plevel=0, pelsearch=3, searchparam=2, search=3, lambda=2000, lsad=1600)
C = S.MRecalculate(A, blksize=8, overlap=2, searchparam=1, search=3)
D = S.MRecalculate(B, blksize=8, overlap=2, searchparam=1, search=3)
V.MSmoothFps(S, C, D, num=60, den=1, algo=13, sadgamma=2, sadml=0, ml=0, blend=true)
GetMTMode(false) > 0 ? distributor() : last

Try it and report back results.
widezu69 is offline  
post #1887 of 1968 Old 02-26-2012, 05:44 AM
Member
 
krneki's Avatar
 
Join Date: Jul 2007
Posts: 17
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
I just installed SVP manager and everything else from full package (madVR, ReClock, etc.). When I tried to play 720p .mkv movie with MPC-HC, SVP kept saying "waiting for ffdshow video" (only ffdshow audio was used during playback), so I added "ffdshow video" into external filter in MPC-HC. But now video is very stuttering, I think graphics card is idle and CPU is on 100 %. Obviously DXVA or CUDA is not working.

How do I force SVP to use DXVA or CUDA? I have core2duo P8400 2.26 GHz and GeForce 9650m GT.

Thanks.
krneki is offline  
post #1888 of 1968 Old 02-28-2012, 08:37 AM
Member
 
Chainik's Avatar
 
Join Date: Apr 2011
Posts: 137
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 15
OK, let's say the end of MVTools2 as we know it has come
I won't do any modifications of MVTools any more, so here it is:

THE FINAL VERSION
With a description, at last

To summarize: the final output is identical to original MVTools 2.5, BUT:
- vectors search (MAnalyse) performance is up to 40% faster
- frame rendering (MSmoothFps) is accelerated with GPU


Let's go forward!
Chainik is offline  
post #1889 of 1968 Old 02-28-2012, 12:49 PM
Member
 
janderclander14's Avatar
 
Join Date: Apr 2006
Posts: 35
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 1 Post(s)
Liked: 13
Quote:
Originally Posted by widezu69 View Post

@tigermaster,

In answer to your first question, the vector recalculation has nothing to do with how many frames you wish to output. Recalculation is basically a second pass. The algorithm scans the changes in the frame, then performs the scan again but also using the vectors from the first pass thus improving frame accuracy. Think of when doing x264 encoding, 2 passes is always better than one even if it takes more time.

The reason why I chose that particular block size is for 1080p is due to the size of the block. The configuration above generally works best for 720p stuff. I'm using 16x16, which when recalculated is using 8x8 in the second pass (that's how it works) overlapping using 0.25 blocks so 4x4 and 2x2 respectively. When using 8x8, that gets reduced to 4x4 the second pass when overlapping, that because 2x2 and 1x1. Unfortunately, the overlap function in MVTools2 does not like 1x1 very much hence I don't recommend using a 8x8 configuration.

720p video uses 16x16 so it makes sense for 1080p, which is 1.5, times larger to use 24x24 but that is not supported so, the next best thing is 32x16 which is the same in terms of surface area.

I assume you are using MPC-HC using ffdshow. Might I suggest that you disable SVP and exit from it completely and try using a script just to see whether it makes any difference.

Basically, play a video, pause, right click and go to Filters, ffdshow video filter, properties and go to the Avisynth tab. Make sure Avisynth, YV12 and Buffer back/ahead are all checked and nothing else. Buffer back/ahead should be 0/24.

Put this script in the area below it:

svp_flow_lib="C:\\Program Files (x86)\\SVP\\Plugins\\MVTools2\\libflowgpu.dll"
svp_scheduler=true
SetMTMode(1,15)
V = ffdshow_source()
SetMTMode(2)
S = V.MSuper(pel=2, hpad=16, vpad=16, rfilter=4)
A = S.MAnalyse(isb=true, blksize=16, overlap=4, plevel=0, pelsearch=3, searchparam=2, search=3, lambda=2000, lsad=1600)
B = S.MAnalyse(isb=false, blksize=16, overlap=4, plevel=0, pelsearch=3, searchparam=2, search=3, lambda=2000, lsad=1600)
C = S.MRecalculate(A, blksize=8, overlap=2, searchparam=1, search=3)
D = S.MRecalculate(B, blksize=8, overlap=2, searchparam=1, search=3)
V.MSmoothFps(S, C, D, num=60, den=1, algo=13, sadgamma=2, sadml=0, ml=0, blend=true)
GetMTMode(false) > 0 ? distributor() : last

Try it and report back results.

I'm currently using a modified version of the above script on an overclocked i7 2600k (4 cores, eight threads, 4,4 Ghz) and outputing it to a 720p projector @50hz:

svp_flow_lib="C:\\Program Files (x86)\\Avisynth 2.5\\plugins\\libflowgpu.dll"
svp_scheduler=true
SetMTMode(1,22)
V = ffdshow_source()
SetMTMode(2,22)
S = V.MSuper(pel=2, hpad=16, vpad=16, rfilter=4)
A = S.MAnalyse(isb=true, blksize=16, overlap=4, plevel=0, pelsearch=3, search=3, lambda=2000, lsad=1600, dct=5)
B = S.MAnalyse(isb=false, blksize=16, overlap=4, plevel=0, pelsearch=3, search=3, lambda=2000, lsad=1600, dct=5)
C = S.MRecalculate(A, blksize=8, overlap=2, search=3, dct=5)
D = S.MRecalculate(B, blksize=8, overlap=2, search=3, dct=5)
V.MSmoothFps(S, C, D, num=50, den=1, algo=13, block=false, sadgamma=2, sadml=0, ml=0, blend=true)
GetMTMode(false) > 0 ? distributor() : last

As I'm only using 720p content, the cpu usage is below 30% (GPU usage is below 20%). Is there any parameter that could be changed to improve the quality considering the available CPU/GPU resources? I prefer minimum artifacts at the cost of perfect smoothness.

Thanks!
janderclander14 is offline  
post #1890 of 1968 Old 02-28-2012, 01:19 PM
Senior Member
 
SubJunk's Avatar
 
Join Date: May 2010
Posts: 327
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
No, janderclander14 those are the best settings IMO. That is what InterFrame uses by default.
SubJunk is offline  
Reply Home Theater Computers

User Tag List

Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off