real.finder's Avisynth Stuff - Page 35

tormento · 31st January 2021, 19:42

PHP Code:


			
SetMemoryMax()

SetFilterMTMode("DEFAULT_MT_MODE", 2)

LoadPlugin("D:\eseguibili\media\DGDecNV\DGDecodeNV.dll")

DGSource("F:\In\1_50 colore venuto dallo spazio, Il\colore.dgi",ct=140,cb=140,cl=0,cr=0)

CompTest24(1)

SMDegrain (tr=4, thSAD=500, refinemotion=false, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true, n16_out=true)

Prefetch(6)

x265 slow --input-depth 16 --dither 4.43 fps

Perhaps the best compromise yet.

real.finder · 31st January 2021, 21:27

Quote:

Originally Posted by tormento

Please explain me why explicitely define prefilter is faster than implicitely use prefilter=4.

in the code I post it will be 8bit for prefilter, and it will be faster as Boulder told you

Quote:

Doing it in 8 bits gives you a nice boost without actually sacrificing quality.

tormento · 1st February 2021, 08:05

Quote:

Originally Posted by real.finder

in the code I post it will be 8bit for prefilter, and it will be faster as Boulder told you

The bold part is the one I need clarification about:

ConvertBits(10).SMDegrain(tr=4, thSAD=500, refinemotion=false, mode=0, contrasharp=false, PreFilter=SMDegrain_prefilters(prefilter=4,a=1,d=1,h=7.0), truemotion=true, plane=4, chroma=true)

tormento · 1st February 2021, 09:43

Quote:

Originally Posted by real.finder

Code:

SMDegrain(tr=4, thSAD=500, refinemotion=false, mode=0, contrasharp=false, PreFilter=ConvertBits(8).SMDegrain_prefilters(prefilter=4,a=1,d=1,h=7.0), truemotion=true, plane=4, chroma=true)

Code:

SMDegrain (tr=4, thSAD=500, refinemotion=false, mode=0, contrasharp=false, PreFilter=ConvertBits(8).SMDegrain_prefilters(prefilter=4,a=1,d=1,h=7.0),, truemotion=true, plane=4, chroma=true, n16_out=true)

actually turned out to be slower than

Code:

SMDegrain (tr=4, thSAD=500, refinemotion=false, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true, n16_out=true)

4.27 fps vs 4.43.

P.S: n16_out=true doesn't need to have ConvertBit(16) in front of SMDegrain, does it?

real.finder · 1st February 2021, 18:49

Quote:

Originally Posted by tormento

The bold part is the one I need clarification about:

ConvertBits(10).SMDegrain(tr=4, thSAD=500, refinemotion=false, mode=0, contrasharp=false, PreFilter=SMDegrain_prefilters(prefilter=4,a=1,d=1,h=7.0), truemotion=true, plane=4, chroma=true)

as the name said it's just prefilter function of SMDegrain

Quote:

Originally Posted by tormento

Code:

SMDegrain (tr=4, thSAD=500, refinemotion=false, mode=0, contrasharp=false, PreFilter=ConvertBits(8).SMDegrain_prefilters(prefilter=4,a=1,d=1,h=7.0),, truemotion=true, plane=4, chroma=true, n16_out=true)

actually turned out to be slower than

Code:

SMDegrain (tr=4, thSAD=500, refinemotion=false, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true, n16_out=true)

4.27 fps vs 4.43.

they should be the same in that case

edit: try add ,device_type="GPU" to SMDegrain_prefilters

anyway there are no point of using n16_out with external PreFilter like this

Quote:

Originally Posted by tormento

P.S: n16_out=true doesn't need to have ConvertBit(16) in front of SMDegrain, does it?

n16_out is like the lsb_out but a bit faster and only work with 8bit input

tormento · 1st February 2021, 19:39

Quote:

Originally Posted by real.finder

as the name said it's just prefilter function of SMDegrain

Ok, but why to use it instead of prefilter=4 in the 8 bit source case?

Quote:

Originally Posted by real.finder

anyway there are no point of using n16_out with external PreFilter like this

Ok, why? I thought n16_out is used to output 16 bit depth video. Am I wrong? What should I use to deliver a 16 bit video out from *ANY* input source?

Quote:

n16_out is like the lsb_out but a bit faster and only work with 8bit input

So if I have a 10/12 bit input, have I to ConvertBits(16) before using SMDegrain?

real.finder · 1st February 2021, 20:19

Quote:

Originally Posted by tormento

Ok, but why to use it instead of prefilter=4 in the 8 bit source case?

who say use it in 8 bit source? you asked for speed optimization and I give you some, no more no less

Quote:

Ok, why? I thought n16_out is used to output 16 bit depth video. Am I wrong? What should I use to deliver a 16 bit video out from *ANY* input source?

it is but it only work for 8bit input, see here https://forum.doom9.org/showthread.p...16#post1837161

if you need 16bit from 10-14 bits then use ConvertBits(16) before SMDegrain

Quote:

So if I have a 10/12 bit input, have I to ConvertBits(16) before using SMDegrain?

it's up to you, SMDegrain can work with those as they are (input depth = output depth)

tormento · 1st February 2021, 20:24

Quote:

Originally Posted by real.finder

if you need 16bit from 10-14 bits then use ConvertBits(16) before SMDegrain

What will happen if I feed 10 bit source without ConvertBits(16) but with n16_out?

real.finder · 1st February 2021, 20:25

Quote:

Originally Posted by tormento

What will happen if I feed 10 bit source without ConvertBits(16) but with n16_out?

you should get an error message

tormento · 2nd February 2021, 17:08

Ok, the results. Bit long and tedious job but it will save me some encoding time.

Code:

Number of frames:                     4560
Length (hh:mm:ss.ms):         00:03:10.190
Frame width:                          1920
Frame height:                         1040
Framerate:                          23.976 (24000/1001)
Colorspace:                           YV12

Base script:

PHP Code:


			
SetMemoryMax()
SetFilterMTMode("DEFAULT_MT_MODE", 2)
LoadPlugin("D:\eseguibili\media\DGDecNV\DGDecodeNV.dll")
DGSource("F:\In\2_37 padrino III coda, Il - La morte di Michael Corleone\coda.dgi",ct=20,cb=20,cl=0,cr=0)
(*)
Prefetch(6)

where (*) are the differences that I will list below.

Code:

[1] SMDegrain (tr=6, thSAD=600, refinemotion=false, n16=true, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true)
[2] ConvertBits(10).SMDegrain (tr=6, thSAD=600, refinemotion=false, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true)
[3] ConvertBits(12).SMDegrain (tr=6, thSAD=600, refinemotion=false, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true)
[4] ConvertBits(16).SMDegrain (tr=6, thSAD=600, refinemotion=false, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true)
[5] SMDegrain (tr=6, thSAD=600, refinemotion=false, n16_out=true, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true)

AVSMeter64 results:

Code:

FPS  min  |   max  | average
[1] 1.238 |  46512 | 10.96
[2] 1.222 |  256.8 |  7.679
[3] 1.174 |  46083 |  6.170
[4] 1.084 | 140845 |  6.210
[5] 1.272 |  35461 | 10.97

x265 encoding:

Code:

[1] x265_x64.exe -D 10 --crf 22.0 --preset slow --profile main10 --level 4.1 --keyint 240 --vbv-bufsize 78125 --vbv-maxrate 62500 --aq-mode 2 --sar 1:1
[2] same as [1]
[3] same as [1] --input-depth 12 --dither
[4] same as [1] --input-depth 16 --dither
[5] same as [1] --input-depth 16 --dither

x265 results:

Code:

[1] 3.51 fps, 1734.92 kb/s, Avg QP:25.95
[2] 3.23 fps, 1713.39 kb/s, Avg QP:25.92
[3] 2.86 fps, 1713.58 kb/s, Avg QP:25.91
[4] 2.75 fps, 1716.02 kb/s, Avg QP:25.91
[5] 3.55 fps, 1714.94 kb/s, Avg QP:25.91

Can't understand the speedup with out16.

real.finder · 2nd February 2021, 17:33

Quote:

Originally Posted by tormento

Ok, the results. Bit long and tedious job but it will save me some encoding time.

Code:

Number of frames:                     4560
Length (hh:mm:ss.ms):         00:03:10.190
Frame width:                          1920
Frame height:                         1040
Framerate:                          23.976 (24000/1001)
Colorspace:                           YV12

Base script:

PHP Code:


			
SetMemoryMax()

SetFilterMTMode("DEFAULT_MT_MODE", 2)

LoadPlugin("D:\eseguibili\media\DGDecNV\DGDecodeNV.dll")

DGSource("F:\In\2_37 padrino III coda, Il - La morte di Michael Corleone\coda.dgi",ct=20,cb=20,cl=0,cr=0)

(*)

Prefetch(6)

where (*) are the differences that I will list below.

Code:

[1] SMDegrain (tr=6, thSAD=600, refinemotion=false, n16=true, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true)
[2] ConvertBits(10).SMDegrain (tr=6, thSAD=600, refinemotion=false, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true)
[3] ConvertBits(12).SMDegrain (tr=6, thSAD=600, refinemotion=false, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true)
[4] ConvertBits(16).SMDegrain (tr=6, thSAD=600, refinemotion=false, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true)
[5] SMDegrain (tr=6, thSAD=600, refinemotion=false, n16_out=true, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true)

AVSMeter64 results:

Code:

FPS  min  |   max  | average
[1] 1.238 |  46512 | 10.96
[2] 1.222 |  256.8 |  7.679
[3] 1.174 |  46083 |  6.170
[4] 1.084 | 140845 |  6.210
[5] 1.272 |  35461 | 10.97

x265 encoding:

Code:

[1] x265_x64.exe -D 10 --crf 22.0 --preset slow --profile main10 --level 4.1 --keyint 240 --vbv-bufsize 78125 --vbv-maxrate 62500 --aq-mode 2 --sar 1:1
[2] same as [1]
[3] same as [1] --input-depth 12 --dither
[4] same as [1] --input-depth 16 --dither
[5] same as [1] --input-depth 16 --dither

x265 results:

Code:

[1] 3.51 fps, 1734.92 kb/s, Avg QP:25.95
[2] 3.23 fps, 1713.39 kb/s, Avg QP:25.92
[3] 2.86 fps, 1713.58 kb/s, Avg QP:25.91
[4] 2.75 fps, 1716.02 kb/s, Avg QP:25.91
[5] 3.55 fps, 1714.94 kb/s, Avg QP:25.91

Can't understand the speedup with out16.

see here https://forum.doom9.org/showthread.p...16#post1835681

also you can speed 2-4 as Boulder told you and I post an example code to do that

tormento · 2nd February 2021, 17:35

Quote:

Originally Posted by real.finder

also you can speed 2-4 as Boulder told you and I post an example code to do that

That is what I am doing, i.e. process everything in 16 bit and dither in x265.

He asked you:

Quote:

Originally Posted by Boulder

Is this utilized in the SMDegrain port?

real.finder · 2nd February 2021, 17:47

Quote:

Originally Posted by tormento

He asked you:

edit: I mean that part of his post

Quote:

pinterf's MVTools has the possibility to use MAnalyse and MRecalculate at a lower bitdepth and do all the processing at a higher bitdepth. In this case it would mean creating the vectors at 8 bits and do all the denoising etc. at 16 bits which makes things quite a bit faster. Is this utilized in the SMDegrain port?

also

Quote:

motion analysis tends to be the slowest part of the whole thing. Doing it in 8 bits gives you a nice boost without actually sacrificing quality.

and I answer

Quote:

Originally Posted by real.finder

you can by using prefilter as clip

then I give how to do that for you

Quote:

Originally Posted by real.finder

Code:

#8bit source input
ConvertBits(XX).SMDegrain(tr=4, thSAD=500, refinemotion=false, mode=0, contrasharp=false, PreFilter=SMDegrain_prefilters(prefilter=4,a=1,d=1,h=7.0,device_type="GPU"), truemotion=true, plane=4, chroma=true)

or if you have HBD source already

Code:

#HBD source input
SMDegrain(tr=4, thSAD=500, refinemotion=false, mode=0, contrasharp=false, PreFilter=ConvertBits(8).SMDegrain_prefilters(prefilter=4,a=1,d=1,h=7.0,device_type="GPU"), truemotion=true, plane=4, chroma=true)

(same old post but with small edit)

Boulder · 2nd February 2021, 17:55

For HDR sources, I recommend ConvertBits(10) instead of ConvertBits(8) for prefilter as the source has information in the whole 10-bit range. At least DGSource will output 16-bit video in such cases (it's what the card just does).

real.finder · 2nd February 2021, 18:05

Quote:

Originally Posted by Boulder

For HDR sources, I recommend ConvertBits(10) instead of ConvertBits(8) for prefilter as the source has information in the whole 10-bit range. At least DGSource will output 16-bit video in such cases (it's what the card just does).

prefilter it's just for Motion Analyse even if your source is 16bit you can use 8bit without loosing quality, but if you care that much about HBD and you have fast CPU and you don't care about time then you can Motion Analyse in HBD and maybe you will get some better Motion Analyse from that in the end

anyway, I don't know about DGSource, but I think LSMASHSource can do HW decode for 10bit without convert it to 16bit

tormento · 2nd February 2021, 18:27

Quote:

Originally Posted by real.finder

same old post but with small edit

The code you are giving me does SMDegrain on 8 bit source giving out 8 bit video. I need 10 bit output (or higher) to feed 10 bit x265 with.

real.finder · 2nd February 2021, 19:02

Quote:

Originally Posted by tormento

The code you are giving me does SMDegrain on 8 bit source giving out 8 bit video. I need 10 bit output (or higher) to feed 10 bit x265 with.

tormento · 2nd February 2021, 19:06

Quote:

Originally Posted by real.finder

I lost the initial conversion

I tried some posts ago and it was slower than regular pre=4

real.finder · 2nd February 2021, 20:57

Quote:

Originally Posted by tormento

I lost the initial conversion

I tried some posts ago and it was slower than regular pre=4

you tried vs n16 things, of course n16 and even n16_out are the fastest high bit method

edit: also it was without device_type="GPU"

I said for your case above

Quote:

Originally Posted by tormento

AVSMeter64 results:

Code:

FPS  min  |   max  | average
[1] 1.238 |  46512 | 10.96
[2] 1.222 |  256.8 |  7.679
[3] 1.174 |  46083 |  6.170
[4] 1.084 | 140845 |  6.210
[5] 1.272 |  35461 | 10.97

for 2, 3 and 4 not for 1 and 5 which use n16 things

tormento · 9th March 2021, 12:08

Launching AVSMeter_x64 -avsinfo, I am getting:

Function duplicates:

"undefined" : "[InternalFunction]"
"Undefined" : "D:\Programmi\Media\AviSynth+\plugins64\Zs_RF_Shared-1.143~realfinder.avsi"

Why?

31st January 2021, 19:42	#681 \| Link
tormento Acid fr0g Join Date: May 2002 Location: Italy Posts: 2,563	PHP Code: `SetMemoryMax() SetFilterMTMode("DEFAULT_MT_MODE", 2) LoadPlugin("D:\eseguibili\media\DGDecNV\DGDecodeNV.dll") DGSource("F:\In\1_50 colore venuto dallo spazio, Il\colore.dgi",ct=140,cb=140,cl=0,cr=0) CompTest24(1) SMDegrain (tr=4, thSAD=500, refinemotion=false, mode=0, contrasharp=false, PreFilter=4, truemotion=true, plane=4, chroma=true, n16_out=true) Prefetch(6)` x265 slow --input-depth 16 --dither 4.43 fps Perhaps the best compromise yet. __________________ @turment on Telegram

2nd February 2021, 17:55	#694 \| Link
Boulder Pig on the wing Join Date: Mar 2002 Location: Finland Posts: 5,729	For HDR sources, I recommend ConvertBits(10) instead of ConvertBits(8) for prefilter as the source has information in the whole 10-bit range. At least DGSource will output 16-bit video in such cases (it's what the card just does). __________________ And if the band you're in starts playing different tunes I'll see you on the dark side of the Moon...

9th March 2021, 12:08	#700 \| Link
tormento Acid fr0g Join Date: May 2002 Location: Italy Posts: 2,563	Launching AVSMeter_x64 -avsinfo, I am getting: Function duplicates: "undefined" : "[InternalFunction]" "Undefined" : "D:\Programmi\Media\AviSynth+\plugins64\Zs_RF_Shared-1.143~realfinder.avsi" Why? __________________ @turment on Telegram

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode