MVTools

[LaTo]

Originally Posted by superuser

which is recommended version to use at present?

If you want compatibly with official mvtools: 1.9.7.1, else 1.9.7.5...

[josey_wells]

Originally Posted by mikeytown2

1.9.7.5 IC10 builds bomb on MVFlowFps. I only have a conroe so I haven't tried the SSE4 dll. TempGaussMC_beta2 & MCBob don't crash in AvsP though. Vdub doesn't crash with the IC10 Builds, but the output isn't correct (blurred output, YUY2 & YV12).

MSVC08-SSE2 seems flawless, so I'll be staying away from the IC10 builds for now.

This was a brand new computer that I performed the builds on. So maybe I missed something setting of the environment. I will try to reperform the builds this weekend and repost to see if it corrected the problems you see with your famous MVFlowFPS script. V1.9.7.4 with rounding fixes had very good PSNR for the script but did not completely fix all the issues.

There is no difference between V1.9.7.5 and V1.9.7.4 except for the fix for the planes issue in MVDegrainXXX which involved two lines of code that I miss copied and pasted.

[Fizick]

"rounding fixes" demystified.

v.1.10.2.X:

nLambda = nLambda*LSAD/(LSAD + (predictor.sad>>1))*LSAD/(LSAD + (predictor.sad>>1));
// replaced hard threshold by soft in v1.10.2 by Fizick (a liitle complex expression to avoid overflow)

v1.9.7.2-1.9.7.5:

nLambda = nLambda*(LSAD/(LSAD + (sad>>1)))*(LSAD/(LSAD + (sad>>1)));
// replaced hard threshold by soft in v1.10.2 by Fizick (a liitle complex expression to avoid overflow)

all integers, comment preserved

PSNR is always better with lambda=0

[josey_wells]

Originally Posted by Fizick

"rounding fixes" demystified.

v.1.10.2.X:

nLambda = nLambda*LSAD/(LSAD + (predictor.sad>>1))*LSAD/(LSAD + (predictor.sad>>1));
// replaced hard threshold by soft in v1.10.2 by Fizick (a liitle complex expression to avoid overflow)

v1.9.7.2-1.9.7.5:

nLambda = nLambda*(LSAD/(LSAD + (sad>>1)))*(LSAD/(LSAD + (sad>>1)));
// replaced hard threshold by soft in v1.10.2 by Fizick (a liitle complex expression to avoid overflow)

all integers, comment preserved

PSNR is always better with lambda=0

This is exactly why I don't want to play these games any more. The two lines of code are actually

    int sad = predictor.sadLuma + predictor.sadChromaU + predictor.sadChromaV;
    nLambda = nLambda*(LSAD/(LSAD + (sad>>1)))*(LSAD/(LSAD + (sad>>1)));

which shows the lambda is not zero.

A lot of the rounding error fixes where in PlaneofBlocks.cpp and Interpolation.cpp.

Because some people have found this useful, I will do new builds as promised over the weekend and then someone else can take over where I left off.

[jeffy]

@josey_wells: Thank you for all the improvements you brought to this library and thank you for your multithreading improvements.
MVDegrainMulti brought a really nice simplification to the MVDegrain family of functions.

@Fizick: Thank you for all the things you've done so far.

[Fizick]

Originally Posted by josey_wells

which shows the lambda is not zero.

are you sure?

check value of integer dividing expression in your added brackets

[josey_wells]

OK, I see rearrangement of brackets has caused a multiply by 0 since den>num. I will fix before doing build.

[Fizick]

I also prepare (almost finish) v1.11.4.X, currently in beta testing stage - i am afraid new bugs of new changes:

MVAnalyse: cut unused compensation memory space from vector clip (mc parameter).
Decreased internal cache size growing.
All MVFlow functions: fixed old bug with pixels for pel=4.
MVAnalyse: added rfilter parameter (smooth method).
MVBlockFps: added YUY2 and pelclip support.
Tiny stability fixes for MT.

New MVBlockFps for simple fps change (almost realtime :-).
MVAnalyse: more smooth method of frame hierarchical level reducing (like ReduceBy2) for smoothing at every level.
MVAnalyse: Added pzero parameter of zero vector cost for more coherent motion vectors (now switched ON by default, set to 0 if you need in old algo).
MVAnalyse: Decreased pelsearch internal clipping from pel to 1 (asked by somebody for speed).
MVFlowFps,MVFlowFps2, MVFlowInter: removed thSAD parameter.
All functions: Removed mmx parameter.

I also consider question to add MVAnalyseMulti (and MVDegrainMulti) functions, but without threading. Is it may be useful?

[mikeytown2]

MVDegrainMulti has simpler syntax IMHO when using the Degrain function. My question is if your going threw all the hassles of adding the Multi functions, why kill off threading; are the changes that big?

Glad to see that pel=4 will work correctly, and sounds like MVBlockFps will give us more toys to pay with in terms of FPS conversions.

[Fizick]

I suppose that StackVertical of several vector clips (and reverse operation) is not very hard hassle.

[Blue_MiSfit]

Is it possible to get some basic documentation thrown into these builds / releases? I forget the syntax and looking it up is proving to be a bear

Suggestions?

~MiSfit

[Boulder]

Originally Posted by josey_wells

OK, I see rearrangement of brackets has caused a multiply by 0 since den>num. I will fix before doing build.

Is the build released anywhere yet?

[josey_wells]

Sorry, I have been extremely busy.

Here are new builds on another machine with fix for brackets.

http://rapidshare.com/files/14844030...9.7.7.rar.html

[maxhondur]

Hi,

I have a question about MVTools, whether or not something is feasable or not. I've searched before but if I'm told to search harder that's fine too, I readily admit that MVTools is often hard for me to understand.

I have an animation dvd, DGdecode tells me it's 97% film, therefore I can avoid telecide and decimate and have a nice looking 23.97fps video. However, some parts of the video were meant to be in 29.97fps due to use of CG.

I can therefore make a video that is either 23.97fps and jerky on CG, or a 29.97fps file that doesn't take advantage of the 97% film result (and while being smooth in CG, is stuttery in all the scenes that were supposed to be the 24fps).

So, here's my convoluted question: Can I use mvtools to compare the motion vectors of the 23.97fps video to the 29.97fps video? In making a final 120fps file, I want most of the video from the 23.97fps file, but on CG scenes where the motion is jerky, I want to use parts from the 29.97fps file.

Yes, practically speaking this is a downright silly approach (vfr, anyone). But, I thought it would be cool to use MVTools in such a way nonetheless.

[superuser]

Originally Posted by josey_wells

Sorry, I have been extremely busy.

Here are new builds on another machine with fix for brackets.

http://rapidshare.com/files/14844030...9.7.7.rar.html

Thnxs a lot. Appreciate ur efforts :thumbup:

with this changes, is r the parameters expected to work as original mvtools? if so will switch to this version for time being.

thnxs once again.

Originally Posted by Fizick

I also prepare (almost finish) v1.11.4.X, currently in beta testing stage - i am afraid new bugs of new changes:
...
...

does this build include enhancements made by Josey?

If not, in case r there plans of merging these two builds?

Appreciate efforts devoted by both of u to enhance this amazing plugin.

Originally Posted by LaTo

If you want compatibly with official mvtools: 1.9.7.1, else 1.9.7.5...

thnxs. it helps.

[Fizick]

Originally Posted by superuser

does this build include enhancements made by Josey?

If not, in case r there plans of merging these two builds?

v1.11.x - no (besides memory fixes already merged to v1.10.X).
"multi" syntax will not implemented as well as internal multitreading.

My plans:
1. To release v1.11.4.2 final or may be v1.10.3 (i have some small trouble with v1.11.4).

2. I work on v2.0 alpha without idx. It (will) have rather big external and internal changes (but i try make them as small as I can )
http://forum.doom9.org/showthread.php?t=131033
I hope v2.0 will have better performance and stability. Welcome to that thread for development discussion!
So, we will have 3 branches.

3. Then some best MT method (methods) may be considered and implemented to mvtools v2.1. I hope avisynth v2.6 will have some progress too.

[Terranigma]

Originally Posted by Fizick

My plans:
1. To release v1.11.4.2 final or may be v1.10.3 (i have some small trouble with v1.11.4).

What kind of trouble and how big of an issue is it?

[Fizick]

Terranigma, i found some small old bug with help of new v2.0. Now it is fixed (not important).
So, MVTools v1.11.4.2 final is released today.

[45tripp]

don;t know if anyone else gets this,
but i get weird results with josey's last build and fft3dgpu prefitering.
basically with fft3dgpu(...,bt=1,2,or 3).mvanalysemulti(...,blksize16)

PHP Code:

a=last
b=a.FFT3Dgpu(bt=3).MVanalyseMulti(refframes=3,blksize=16,overlap=8,idx=1)
a.MVDegrainMulti(b,refframes=3,thSAD=600,idx=2) 


tripp

[Sagekilla]

Dear god, how many branches do we have? Josey_wells has one, then there's the 1.11.x and 1.10.x and 2.0!

What's the difference of 1.11 and 1.10?