lsy17096535
帖子: 2
注册时间: 2015-10-29 16:24
联系: Skype

Re: BM3D-r3 最佳的降噪算法之一,BM3D和V-BM3D的VS滤镜实现

我用了,这个输入应该是什么?YUV420可以吗?或者avi什么的?我明天贴一下我的脚本,帮我看一下行吗? 谢谢
InvisibleArts
帖子: 8
注册时间: 2015-07-15 21:40

Re: BM3D-r3 最佳的降噪算法之一,BM3D和V-BM3D的VS滤镜实现

lsy17096535 写了:我用了,这个输入应该是什么?YUV420可以吗?或者avi什么的?我明天贴一下我的脚本,帮我看一下行吗? 谢谢
RTFM. (Copied from the script)
################################################################################################################################
## Main function: BM3D()
################################################################################################################################
## A wrap function for BM3D/V-BM3D denoising filter
## The BM3D filtering is always done in 16-bit int or 32-bit float opponent(OPP) color space internally.
## It can automatically convert any input color space to OPP and convert it back after filtering.
## Alternatively, you can specify "output" to force outputting RGB or OPP, and "css" to change chroma subsampling.
## For Gray input, no color space conversion is involved, thus "output" and "css" won't take effect.
## You can specify "refine" for any number of final estimate refinements.
################################################################################################################################
## Basic parameters
## input {clip}: clip to be filtered
## can be of YUV/RGB/Gray/YCoCg color family, can be of 8-16 bit integer or 16/32 bit float
## sigma {float[]}: same as "sigma" in BM3D, used for both basic estimate and final estimate
## the strength of filtering, should be carefully adjusted according to the noise
## set 0 to disable the filtering of corresponding plane
## default: [5.0,5.0,5.0]
## radius1 {int}: temporal radius of basic estimate
## - 0: BM3D (spatial denoising)
## - 1~16: temporal radius of V-BM3D (spatial-temporal denoising)
## default: 0
## radius2 {int}: temporal radius of final estimate
## default is the same as "radius1"
## profile1 {str}: same as "profile" in BM3D basic estimate
## default: "fast"
## profile2 {str}: same as "profile" in BM3D final estimate
## default is the same as "profile1"
################################################################################################################################
## Advanced parameters
## refine {int}: refinement times
## - 0: basic estimate only
## - 1: basic estimate + refined with final estimate, the original behavior of BM3D
## - n: basic estimate + refined with final estimate for n times
## each final estimate takes the filtered clip in previous estimate as reference clip to filter the input clip
## default: 1
## pre {clip} (optional): pre-filtered clip for basic estimate
## must be of the same format and dimension as the input clip
## should be a clip better suited for block-matching than the input clip
## ref {clip} (optional): basic estimate clip
## must be of the same format and dimension as the input clip
## replace the basic estimate of BM3D and serve as the reference clip for final estimate
## psample {int}: internal processed precision
## - 0: 16-bit integer, less accuracy, less memory consumption
## - 1: 32-bit float, more accuracy, more memory consumption
## default: 0
################################################################################################################################
## Parameters of input properties
## matrix {int|str}: color matrix of input clip, only makes sense for YUV/YCoCg input
## check GetMatrix() for available values
## default: guessed according to the color family and size of input clip
## full {bool}: define if input clip is of full range
## default: guessed according to the color family of input clip and "matrix"
################################################################################################################################
## Parameters of output properties
## output {int}: type of output clip, doesn't make sense for Gray input
## - 0: same format as input clip
## - 1: full range RGB (converted from input clip)
## - 2: full range OPP (converted from full range RGB, the color space where the filtering takes place)
## default: 0
## css {str}: chroma subsampling of output clip, only valid when output=0 and input clip is YUV/YCoCg
## check ToYUV() for available values
## default is the same as that of the input clip
## depth {int}: bit depth of output clip, can be 8-16 for integer or 16/32 for float
## default is the same as that of the input clip
## sample {int}: sample type of output clip, can be 0(vs.INTEGER) or 1(vs.FLOAT)
## default is the same as that of the input clip
################################################################################################################################
## Parameters of depth conversion
## dither, useZ, ampo, ampn, dyn, staticnoise:
## same as those in Depth()
################################################################################################################################
## Parameters of resampling
## cu_kernel, cu_taps, cu_a1, cu_a2, cu_cplace:
## used for chroma up-sampling, same as those in fmtc.resample
## default: kernel="bicubic", a1=0, a2=0.5, also known as "Catmull-Rom"
## cd_kernel, cd_taps, cd_a1, cd_a2, cd_cplace:
## used for chroma down-sampling, same as those in fmtc.resample
## default: kernel="bicubic", a1=0, a2=0.5, also known as "Catmull-Rom"
################################################################################################################################
## Parameters of BM3D basic estimate
## block_size1, block_step1, group_size1, bm_range1, bm_step1, ps_num1, ps_range1, ps_step1, th_mse1, hard_thr:
## same as those in bm3d.Basic/bm3d.VBasic
################################################################################################################################
## Parameters of BM3D final estimate
## block_size2, block_step2, group_size2, bm_range2, bm_step2, ps_num2, ps_range2, ps_step2, th_mse2:
## same as those in bm3d.Final/bm3d.VFinal
################################################################################################################################
Echo
帖子: 7
注册时间: 2016-08-09 18:35

Re: BM3D-r3 最佳的降噪算法之一,BM3D和V-BM3D的VS滤镜实现

mawen1250 写了:
lsy17096535 写了:斑竹你好 能发个bm3d处理的脚本吗?我这里一直报错,最近做压片,想提高一下亮度,并且降噪,谢谢啦。
用mvsfunc里BM3D
请教版主,用了mvsfunc里BM3D(脚本如下),提示错误Python exception: No attribute with the name fmtc exists
import mvsfunc as mvf
core.max_cache_size = 4000
dst = mvf.BM3D(src, sigma=[3,3,3], radius1=0)
/*********************************/
看了一下mvsfunc脚本,其中BM3D的Depth、ToRGB 、ToYUV中用到了fmtc
查了好几遍,感觉应该是包、插件 加载异常的问题,但未找到问题根本
新手上路,可否指点,非常感谢
wzwenyuyu
帖子: 12
注册时间: 2016-01-03 1:50

Re: BM3D-r3 最佳的降噪算法之一,BM3D和V-BM3D的VS滤镜实现

Echo 写了:
mawen1250 写了:
lsy17096535 写了:斑竹你好 能发个bm3d处理的脚本吗?我这里一直报错,最近做压片,想提高一下亮度,并且降噪,谢谢啦。
用mvsfunc里BM3D
请教版主,用了mvsfunc里BM3D(脚本如下),提示错误Python exception: No attribute with the name fmtc exists
import mvsfunc as mvf
core.max_cache_size = 4000
dst = mvf.BM3D(src, sigma=[3,3,3], radius1=0)
/*********************************/
看了一下mvsfunc脚本,其中BM3D的Depth、ToRGB 、ToYUV中用到了fmtc
查了好几遍,感觉应该是包、插件 加载异常的问题,但未找到问题根本
新手上路,可否指点,非常感谢
fmtconv
另外请先查阅http://www.vapoursynth.com/doc/
fch1993
帖子: 213
注册时间: 2012-06-12 11:56

Re: BM3D-r6 | BM3D和V-BM3D降噪算法的VS滤镜实现

自己根据最新源码+fftw 3.3.5编译了一个新版本,编译设置其他和官方版一致。

下载地址2选1。

https://mega.nz/#!ldZnjDoD!9Qq4E7y528U0 ... SlP79KIw74
https://drive.google.com/file/d/0B2tAdv ... sp=sharing

mvf.BM3D(src, sigma=3.0, profile1="fast")
这个测试
相比r6+fftw 3.3.4,提速大概有10%。(8线程下,原版2.10fps,新版2.37fps)。
考虑fftw 3.3.5支持AVX2指令集,可能在新CPU下会有更多加成。

PS:貌似3.3.5对于ICL不太友好,我自己编译的无论是BM3D还是dfftest,按照holy的配置下去结果都是比MSVC略慢(ICL还是O3的情况下)。
Echo
帖子: 7
注册时间: 2016-08-09 18:35

Re: BM3D-r3 最佳的降噪算法之一,BM3D和V-BM3D的VS滤镜实现

wzwenyuyu 写了:
Echo 写了: 提示错误Python exception: No attribute with the name fmtc exists
看了一下mvsfunc脚本,其中BM3D的Depth、ToRGB 、ToYUV中用到了fmtc
fmtconv
另外请先查阅http://www.vapoursynth.com/doc/
感谢解惑,已完成vpy,可正常preview,非常感谢。
目前在用vspipe做最后的压片、视频保存,不太顺利,继续加油吧O(∩_∩)O~

回到 “VapourSynth”