SVP 插帧 vs 原始视频 vs Pot（非 ffdshow) 内建倍帧 同屏对比[阿凡达打斗场景 146MB]

ligang989xx

POT的错误不是一般的多，主要问题还是移动条纹撕裂，这是致命的缺陷。普通的画面还好一但遇到左右移动的画面就会产生破碎。

morning44

whindsonh 发表于 2011-7-31 20:02
这个....破解4核后FX5Koc2.7+GTS250跑1080p没问题的说.....
测试视频是TRON,这个比阿还耗cpu的说.....

兄弟x215+集成主板软解任何高清的说，感觉这个插帧不就是和硬解差不多啊

morning44

我个人觉得这个没必要搞什么，，，插帧，尽管我配置足够。一个x215 c68 软解任何高清，硬解也不错
一个四核+5850   就硬解软解看看够了

whateveru

LS你后悔了还是可以皈依主......

结果

屏摄视频续期了
难道4核真的都招架不住“Pot破碎拳”？求真相……
http://u.115.com/file/dn6u7a6r

FullHD8

不知道Pot的效果能比的上这段代码么？求高人指点！

1. SuperRegular = MSuper(hpad=16, vpad=16, rfilter=4)
   
2. 
   
3.         prefiltered = DeGrainMedian()
   
4. 
   
5.                 CustomPelClip = nnedi3_rpow2(4, nsize=3, nns=4, cshift="Spline144Resize")
   
6. 
   
7.                 Super = MSuper(prefiltered, hpad=16, vpad=16, rfilter=4, pel=4, pelclip=CustomPelClip)
   
8. 
   
9. 
   
10.                 backward = MAnalyse(Super, isb=true , blksize=16, blksizev=16, search=3, searchparam=16, pelsearch=16, overlap=4, overlapv=4, truemotion=true, plevel=2, lambda=2000, lsad=1600, dct=0)
    
11.                 forward = MAnalyse(Super, isb=false, blksize=16, blksizev=16, search=3, searchparam=16, pelsearch=16, overlap=4, overlapv=4, truemotion=true, plevel=2, lambda=2000, lsad=1600, dct=0)
    
12. 
    
13.                 backward_2 = MRecalculate(Super, backward, blksize=16, blksizev=8, search=3, searchparam=16, overlap=4, overlapv=2, truemotion=true, lambda=1000, dct=0)
    
14.                 forward_2 = MRecalculate(Super, forward, blksize=16, blksizev=8, search=3, searchparam=16, overlap=4, overlapv=2, truemotion=true, lambda=1000, dct=0)
    
15. 
    
16.                 backward_3 = MRecalculate(Super, backward_2, blksize=8, blksizev=8, search=3, searchparam=16, overlap=2, overlapv=2, truemotion=true, lambda=500, dct=4)
    
17.                 forward_3 = MRecalculate(Super, forward_2, blksize=8, blksizev=8, search=3, searchparam=16, overlap=2, overlapv=2, truemotion=true, lambda=500, dct=4)
    
18. 
    
19.                 backward_4 = MRecalculate(Super, backward_3, blksize=8, blksizev=4, search=3, searchparam=16, overlap=2, overlapv=0, truemotion=true, lambda=250, dct=0)
    
20.                 forward_4 = MRecalculate(Super, forward_3, blksize=8, blksizev=4, search=3, searchparam=16, overlap=2, overlapv=0, truemotion=true, lambda=250, dct=0)
    
21. 
    
22.                 backward_5 = MRecalculate(Super, backward_4, blksize=4, blksizev=4, search=3, searchparam=16, overlap=0, overlapv=0, truemotion=true, lambda=125, dct=4)
    
23.                 forward_5 = MRecalculate(Super, forward_4, blksize=4, blksizev=4, search=3, searchparam=16, overlap=0, overlapv=0, truemotion=true, lambda=125, dct=4)
    
24. 
    
25. 
    
26.                 svp_flow_lib = "C:\Program Files (x86)\AviSynth 2.5\plugins\libflowgpu.dll"
    
27.                 Finest = MFinest(SuperRegular)
    
28. 
    
29.         MSmoothFps(SuperRegular, backward_5, forward_5, finest=Finest, num=240, den=1, algo=13, ml=100, sadml=255, sadgamma=2, block=true, blend=false)
    
30. 
    
31. BlendFPS(60)
    

复制代码

需要具备NNEDI3.dll、SplineResize.dll以及Motion.dll谢谢！

fshx

结果 发表于 2011-9-17 16:47
屏摄视频续期了
难道4核真的都招架不住“Pot破碎拳”？求真相……
http://u.115.com/file/dn6u7a6r

到期了……

fdavid22

SVP+LAV CUVID+madvr dxva 自家Q9400+560ti好舒服啊，就是cpu占用70-90！！基本设定svp中高设定 ！就是人物快动的时候边缘有模糊，估计运算真的没这么强的吧，希望以后改进，低运动的画面完美，看多很多东西，不用忍受那些跳帧的顿挫感了，内挂字幕还是残念，直接射手找外挂字幕搞！！ps：EIZOl887 色彩不是一般的好，其他强人估计更舒服，准备拿去客厅G20C等离子试试实践效果咯！

sikele2236

求教怎么插桢转码