RV9/10 and etc.Hybrid Material encoding for Animes (2nd stage)

[lunaticmoon]

(To 2nd stage, following)
http://forum.doom9.org/showthread.php?s=&postid=529395#post529395

Hail.

First, sorry for my bad English sence and sentences.

Recently, I am using to make RV10 for mixed fps anime clips.
Current most anime clips have this mixed fps feature.
With RV10, I tried three approaches to show those clips smoothly.
But, all of them have no effect on those clips showing smoothly.

I am hopeful more effective optimization in RV10.
Some of my friend think so too.

In world wide ? i dont know yet. ;)

1) Use DropeDupe

DropDupe drops some same continuous frames.
I think this is effective for 24fps/30fps mixed clips, but not so.

ex. 120fps approach on 30fps to 24fps seems
[1/30][1/30][1/30][1/30][1/30]
-> [1/24 ][1/24 ][1/24 ][1/24 ]�@�@

ex. DropDupe approach on 30fps to 24fps seems
[1/30][1/30][1/30][1/30][1/30]
-> [1/30][1/30][1/30][ 2/30 ]

DropDupe seems only effective for streaming.

2) Use Deinterlace/IVTC Auto

Handing over interlace source to RV10
and use Deinterlace/IVTC Auto.

RV10 seems to judge deinterlace methods :
deinterlace(to 30fps?) or IVTC(to 24fps) automatically.

But RV10 dont senses interlace stripes so good.

3) Use interlace maintenance approach

In interlace clip, fps mixed features are not problem,
but stripes showed on speedy scene.

Unfortunately,
RV10 decoder seems to have no progressive decode mode.

Recently, increasing more and more animes using mixed fps feature.

As it stands, my friends use 120fps approach with MPEG4 codecs,
or interlace maintenance mode with WMV9.
Those are fine, but those processes become complicated.

thanks for reading. :)

[Dark-Cracker]

hi,

first welcome on this forum :)

i am a non expert in anime encoding but i have 5 min so il will give u my opinion :) (u have no choice :) ).

first RV10 max FPS is 60 (so u can forget 120 fps encoding method :) ).

i think u should speak about hybrid material instead of mixed clips , this will help u to find information on this forum).

about (2) : why not try decomb it can give better result in hybrid mode than usual IVTC filters. else i think u can try a bob filter with double framerate the payback is generaly more smooth during travelings.

Bye.

[Sirber]

There is also dropdupe filter :)

[lunaticmoon]

Thanks for responses ! :)

In this forum, my thema is called "hybrid material",
I understood.

Serched that words and read AutoRV10 thread,
and found Decomb "APPENDIX B. Dealing with Hybrid Clips".

That seems like DEC60 avisynth filter (made in Japan),
but DEC60 outputs an AVI file.
That means output is constant frame rate, doesnt it ?
also Decomb too ?

DropDupe only drops frame and just in front of frame showed long.
Thas is not smoothly, I think. frame showing time is not correct.

but when I looks the clips, that is not anxious so much.
my eyes are not useful...:(

[Dark-Cracker]

hi,

yes .avi file is CFR (constant frame rate) u can however search on this forum there is a way to made VFR (variable frame rate) but in matroska container the step to use it is a bit complicated (if i right remember).

Decomb and DEC60 work approximatively (<= hum does this word exist ?) the same way. but if i right remember dec60 was used in 2 pass mode , first it generate a .idx file and after it filtered the material.

if u are interested SmartDecimate give also good result on hybrid anime.

>DropDupe only drops frame and just in front of frame showed long.
Thas is not smoothly, I think.

u can adjust threshold to have a better detection of duplicated frames.

Bye.

[lunaticmoon]

Hail.

Thanks for advices. :)

I think Decomb decides at first 30 or 24 basing and blend for the other rate frame.
but blend approach brings worse quality a bit.

In my belonging community, people using RV10 are use approach below:

1. cutting clip by 24fps parts and 30fps parts each other.
2. So many of clip parts created.
3. encode each parts by each fps by RV10
(24fps parts by 24fps encoding, 30fps parts by 30fps encoding)
4. Concatinate all created RMVB by rmeditor.
5. 1 VFR clip made.
6. they sleep in tired :)

their clips have HiQuality view but it too conplicate.
I want more easy approach, and to get their quality.

In week end, i want to try that SmartDecimate. but that is also for CFR ?

I think CFR frame drop approaches results are:

[24][24][24][24][null]/5sec

and true 24fps is:

[24][24][24][24]/5sec

On stats info of MPC Player, displays 24fps both.
but true view timing seems sliped.
I want no null timing frame approach.

Is that exist ? (if possible easy method)

thanks :)

[karl_lillevold]

lunaticmoon: This sounds interesting, but I am afraid it is not entirely clear to me exactly what the problem is. Could you please try to explain the following:

1) how these mixed sequences are structured? (fps, interlace, progressive etc). Why are they structured like this? Could you provide a short example? (it does not have to be that short either, I can provide upload location in Private Message)

2) how would you like them to be encoded?

[lunaticmoon]

Hail.
thanks for resposes.

In the other thread Mr.Karl is interested in the Deinterlace/IVTC Auto mostly, I think.

therefore first, I pick up the three files in this time.
( im sorry private message option, I cant know the usage yet :( )

Those files are uploaded my reantal web (contract piece soon)
http://64.91.238.115/~cyberlan/dearkarl/deinterlaceauto/

deintauto.log : Producer log of RealBatch in Deinterlace/IVTC Auto mode

deintauto.rmvb : fainal output (no audio). In high motion anime part, some combings remain. (I renamed this file to upload therefore no match with the name in log)

interlaced.avs : avisynth2.5 script to maintain interlace.

I think, searching combings ability is not so good ... :(

If you are interest in DROP or NULL frame VS true 24fps,
I write later again. (but I can explains well ? :( )

Thanks for reading :)

[karl_lillevold]

lunaticmoon: Thanks for uploading those files. It shows that auto-IVTC and de-interlace did not quite work for this source. I am not sure why, the pattern maybe changes too quickly. If you can find a way to upload part of your source (see PM, you can use a tool to cut the source file), I will see if I can repro, and file a bug report. I can not promise when we will have time to fix though.

It is also interesting that our de-interlacer / inverse-telecine filters are completely open source, and anyone interested enough can actually get the source, and try to improve themselves. I am not saying this is easy though, the IVTC is rather complicated statistical analysis, and was written by someone who is no longer at Real.

From your log file I can see that you also disabled VideoDupFrameDropper. This is good, because IVTC needs to see all the frames.

Full interlace support throughout the system is also being considered, but is not there now, as you have noticed.

[lunaticmoon]

Hail. :)

Today, I try to show my image of Hybrid Material with a tiny diagram.

That was drawed with M$ Excel 2000.
http://64.91.238.115/~cyberlan/dearkarl/hybridmaterial/hybridmaterial.xls

That diagram's square means the progressive farme(s).

I think 1)'s sequence is most better for VFR container.
30fps frame and 24fps frame have own best timing.

2) is 120fps approach of CFR container that simuklates VFR with dummy frames.
that frames have effective dummy frames to show valid frames on best timing.
dummy frames are only stagger timing and have no valid images.

3) is DEC60() of avisynth filter (likely Decimate()) approach.
DEC60() is not standalone filter. First, FPSCHK.exe judge that any 5 farmes are 30fps or 24fps and make *.idx file written that pattern.
DEC60() read *.idx file and thins out the most suitable frame from 24fps parts, and thind out frames are marked.
Lastly, AVI that DEC60() made hands over to AVI60.exe, and 120fps CFR AVI file are completed.

4) is my image of DropDupe? approach
this approach looks like 3), but on final point, DropDupe? drops dup frame and to make up for timing previous frame occupies droped farame's timing. (But this is my guess and may be not truth. :D )
Some viwer apeal to me that my clips using dropdupe are not smooth.
Therefore, I considered this illusion :D

I want to get smooth linke 1) with dropdupe :(

thanks for reading my illusion :)

TO: Mr.Karl
I want to send you recent source ( cutted and compressed )
but that is quite big yet. I send you PM later about that.

[karl_lillevold]

lunaticmoon: Thanks for the explanation. I understand that in these 120 fps clips, the 24fps parts are already correctly inverse-telecined. It seems to me the VideoDupFrameDropper should have worked to decimate from 120 fps to correct VFR hybrid 24 fps / 30 fps, if Producer had accepted 120 fps video. I am not sure why it does not result in smooth video after the DEC60() process before encoding. Could you also upload a couple of minutes of 120 fps video? There might be a limit at 60 fps in Producer, but I will look into what can be done anyway.

[lunaticmoon]

Hail.

I used some 120fps approach filters.

1) DEC60() avisynth filter and AVI60.exe

http://www.geocities.co.jp/SiliconValley-Sunnyvale/3109/
http://www.geocities.co.jp/SiliconValley-Sunnyvale/3109/oldtools.html

As a result, some combings remain on recently source.:(

2) itvfr.auf, vfrout.auo AVIUTL filter

http://www.tenchi.ne.jp/~yoko/aviutl/vfrout/

As a result, some combings remain on recently source, too.:(

3) Auto60.auo AVIUTL filter

No WebPage

As a result, some combings remain on recently source, too.:(

...

For recently source, 3)'s approach is best, and 2)'s approarch is better, I think. (in default settings)

It seems that the combing judging with any automatic approach is very difficult.It may be said that RV10 also has Hardware Deinterlacer like XviD.

thanks.:)

TO: Mr.Karl
I sent the result files with above approaches, please check PM.

[bond]

sorry for not being totally on topic, but did you try to encode a hybrid source in dshow (important as dshow handles vfr) with a dshow encoder filter (for example the one from 3ivx or nerodigital), together with outputting to .mp4 (which is able to handle vfr streams)?

if this would work it would be a pretty easy and straight forward approach for encoding mixed fps material (aka vfr material)


btw its already possible to remux from 120fps .avis to .mp4 with the 3ivx mp4 muxer and getting a vfr mpeg-4 stream inside the .mp4 container without the need for reencoding (it should be also possible with the mp4box mp4 muxer, but i never tried it)

[lunaticmoon]

thanks bond for response.

Honestly, I have not understood the contents of your report in detail.
It is because the skill of my English is lacking yet.
Can the resource of 3ivx or NeroDigital use a free thing?
Does the technique prevail globally again?
I think that time will be found and he will also investigate.

thanks.:)

[bond]

Originally posted by lunaticmoon
Can the resource of 3ivx or NeroDigital use a free thing?what do you mean?
3ivx is a 100% free mpeg-4 encoder (no adware), you can encode in graphedit any input which you can play in a dshow based player (like a dvd)

Does the technique prevail globally again?3ivx is a mpeg-4 encoder, as xvid and divx5

the container format defined in the mpeg-4 standard is .mp4, which in contrary to .avi can handle variable framerate, which means there is no need for these hybrid 120fps .avis

try the following
1) open the .vob in graphedit
2) remove anything but the .vob itself and the mpeg-2 demultiplexer
3) connect the 3ivx video encoder to it and set the encoding settings you want
4) connect the 3ivx encoder to the 3ivx mp4 muxer (dont change any settings in the muxer)
5) connect the 3ivx muxer to the file writer (use the .mp4 extension)
6) push play and if everything goes right the output is a variable framerate mpeg-4 video stream inside the .mp4 container

[lunaticmoon]

Although possibly my how to write was bad, I cannot enjoy a metaphor in English yet in my command of English.(^^;

thanks Mr.Bond for explanation of a detailed procedure.:)
I want to try that procedure on next weekend.

btw, May I check a few ?

1) Can AVI contain the VFR material ?
2) If your evaluation is right, 3ivx VFR approach is nice to recently Animes. but it seems that the subject of 3ivx is not rising so much. Why ?

thanks for reading and responses.:)

[bond]

Originally posted by lunaticmoon
1) Can AVI contain the VFR material ?no, only smart containers, like .mp4 or matroska can store vfr. talking about matroska, i dunno if the matroska dshow muxer filter is able to handle vfr content (for example from the 3ivx dshow encoder), but the 3ivx mp4 muxer filter should be able to do it
still this has to be tested :)

2) If your evaluation is right, 3ivx VFR approach is nice to recently Animes. but it seems that the subject of 3ivx is not rising so much. Why ?hm, because 3ivx is not so well known till now, and also the .mp4 container isnt (altough its very powerfull and defined in the mpeg-4 standard :D )

still this whole thing is more a codec interface/container issue, meaning the old vfw/avi (as used in virtualdub) is simply not able to handle new technologies like variable framerates, but newer ones, like directshow are

so for example if real would release a dshow encoder it should also work, but the point is it doesnt exist, so you have to try it with 3ivx :)

[Sirber]

Originally posted by bond
no, only smart containers, like .mp4 or matroska can store vfr. talking about matroska, i dunno if the matroska dshow muxer filter is able to handle vfr content (for example from the 3ivx dshow encoder), but the 3ivx mp4 muxer filter should be able to do it
still this has to be tested :) Works with RV10 VFR :)

[bond]

of course it works with rv, but the question was to make it work in an easy way. thats why i proposed using directshow, for which no rv encoder exists afaik

[karl_lillevold]

lunaticmoon was kind enough to send me various examples on how to get from 29.97 mixed interlaced / telecine to 120 fps. Quite a complex tool chain. It would seem that if the auto-deinterlace / IVTC filter in Producer had been able to properly remove all combings, and do IVTC correctly most of the time on this content, it would be a very good solution, and simplify things considerably ..

[lunaticmoon]

I failed.:(

The OP of Anime source the other day is 24fps fixation, and it became clear that it is not Hybrid yesterday. (but It is Hybrid if CM is included :D) . I recognize it as that sample having not been suitable.
On the other hand, it turns out that very Hybrid OP exists. with the help of Mr.satan who have more detailed knowledge than I on a 120fps trend.
It is OP of "Ai yori aoshi -Enishi-".
Also fortunately, that program is under re-broadcast in the area in which I live, and I think that I can capture it. If a new sample goes into a hand, I will re-work.:)

sorry for my misstake.
thanks.

[karl_lillevold]

lunaticmoon: I was able to encode your sample pretty well with options '-di di'. When I used '-di auto', some combings were left like in your encode, with with both ITVC and de-interlace enabled, it looked good. I will send you a PM with the location of my encode. I still have to look into how the two interact, when both are enabled, but I think interlace is applied only when ITVC has failed.

[stephanV]

Originally posted by bond
of course it works with rv, but the question was to make it work in an easy way. thats why i proposed using directshow, for which no rv encoder exists afaik

IIRC you can mux VFR RV in matroska (with graphedit).

[lunaticmoon]

karl_lillevold:
Your 'di' mode video have no combing, and smooth, as long as I saw.:)
I want to also ask an opinion to Mr.satan which is my cooperator.

What option is it? Is it Producer? (Option '-di di')
If it is it, although I will want to read the manual of Producer, is it on the Web? English or Japanese is sufficient as it.

btw, I also sent you a video of manual 120fps approach.
That is made by Mr.satan. (Although I was sorry to the believer of pious Christianity, this handle does not have the reasons of religion and he did not necessarily attach it. Please think that it is a Japanese Roman alphabet.)

Although IVTC or that judgment which is not so is difficult as you say, if it judges visually and divides manually, a mistake will decrease rather than automatic processing.
However, many do not have so much the people with the skill judged visually including I confidence.

TO Helix team members:
My ideal is getting the Encoding environment in which such a high automatic judging and a high automatic output of accuracy.
If there is it, Encoding will be more in fashion.
If RV10 has a more exact judgment function, video of True VFR (in recently diagram) can be got more easily. (3ivx is not tried yet (^^; )

If another method is taken, for example, YSRealAnime uses ITVFR by avisynth (source is opened perhaps with GPL), and if the portion which hits VFROUT(also source is opened with ???) is mounted, I will think that realization will be approached.It introduced recently about ITVFR and VFROUT. These are delivering judgment information using the shared memory.

thanks for reading.;)

[karl_lillevold]

The option I used is '-di di' from the producer command line. Documentation is available here:
https://producerapps.helixcommunity.org/cmdproducer/docs/CommandLine.htm

di - Apply both Deinterlace and Inverse Telecine

You can download the Helix Producer command line from the binary download section on helixcommunity.org (https://helixcommunity.org/beula/download/).

In Easy RealMedia Producer, I think you get the same effect by checking "Deinterlace/Inverse", then uncheck "Automatic" and check "Deinterlace" and "Inverse".

EDIT: Since I did not work IVTC or de-interlace myself and do not know the details on how these options work together, I ran another experiment, and I found that that the option 'di' always sets de-interlace on, even it detects the telecine pattern and correctly applies IVTC. This will reduce video quality. It seems the only option is "auto" and that this needs some improvement to remove combings when the telecine pattern changes too quickly.

This experiment also made me realize a bug in the new rate control... When Producer's inverse-telecine is used (and not decomb in Avisynth), the codec does not get the right framerate, which is needed for accurate rate control. So if you use Producer's inverse-telecine, and the new rate control, it is necessary to tell the codec the framerate, or use the target filesize option. Some GUIs might already do this.

[lunaticmoon]

I could understand about 'di=di' option.
thanks for explanation.

Although I thought that this would resemble this processing in Avisynth, how is it?


ITVFR(ref="TOP",diMode=0,Debug=false)
Auto24fps(false,24,16,"blend",false)

[lunaticmoon]

I will capture a sample there, since Ai yori aoshi-enishi- is re-broadcast at midnight on next Sunday.

To: bond
Also with the misfortune, as for me, installation of 3ivx knew that it might have a bad influence at the Windows environment. I give up the test of 3ivx. :(

To: karl_lillevold
"-di=di" Although video of the option version was the passing mark, it was if the behavior after a scene change has a detailed mistake. (It is the opinion of Mr.satan. I have not noticed it. ^^; )
It seems that the 120fps processed by ITVFR are perfectly nearer as a result at a merely detailed point.
I think whether infanticide logic of ITVFR is mounted in Producer. The output as VFR which does not have the NULL frame, of course is desirable.
Probably the source of ITVFR itself has a GPL license for the improvement version of IT(). Although there should just be no license problem in the inclusion to Producer, how is it?

ITVFR (also included source)
http://www.tenchi.ne.jp/~yoko/aviutl/vfrout/

[bond]

Originally posted by lunaticmoon
To: bond
Also with the misfortune, as for me, installation of 3ivx knew that it might have a bad influence at the Windows environment. I give up the test of 3ivx.sad to hear, what was the problem? i never had one with 3ivx

[lunaticmoon]

Originally posted by bond
sad to hear, what was the problem? i never had one with 3ivx

Although I do not know well, when the worst, re-installation of Windows may be needed. Is Windows brittle OS?

By the way,
To this Sunday, it is me. OP of "Ai yori aoshi -Enishi-" was able to be captured. I created the ITVFR version, the Auto60 version, the -di=auto version, and the -di=di version.

http://64.91.238.115/~cyberlan/dearkarl/enishi/

EnichiOP_auto60.rar : Auto60 filter 120fps approach
EnishiOP_itvfr.rar : ITVFER filter 120fps approach
EnishiOP_rmvbs.rar : RV10 -di option true VFR approaches (with ERMP)

Combing did not come out with these samples. (It is not visible to me at least.) It will be because there was no high motion scenes ?
These samples must be Hybrid Material.(It judged from the log.)

I thought that it would be a subject for RV10, The direction of behaviors on High motion scenes than that of Hybrid Material.
However, it is if it investigates in detail. The treatment of Hybrid Material may not be enough, either.

Hybrid highness motion scenes material on Anime needs to be encoded for the examination of whole volume (ex. Gundam SEED is very very hybrid Material :D), it cannot carry out freely.

I will send source at this time to mr.Karl later ,by the example.

thanks for reading.;)

[lunaticmoon]

This is a follow-up.

I had Mr.satan(Since it is weak to write English as for him, though regrettable, although this place cannot be called directly) identify video which I created.

The passing mark was not able to be got as a result. orz

His opinion is severe as follows.



EnichiOP_auto60 (1m39s 640x480 XviD 119.88fps).avi
There is no problem generally.
But, in 34s-38s,58s-1m6s, some jaggies are in an outline.
The judgment is considerably wrong if a log is seen.

EnishiOP_auto (1m39s).rmvb
At 29s, the petal has combing.
About the half of 30fps portions, some jaggies are in an outline.

EnishiOP_di (1m39s).rmvb
Same as the above.

EnishiOP_itvfr (1m39s 640x480 XviD 119.88fps).avi
I think that it is the best in this time.
I hear that what it must merely be careful of that ITVFR is thoroughly weak to sauce with tele cinema dirty.(IT() is so.)



As for whether it is [ of how much ] comparatively, those whom I can distinguish so far do not attach imagination.
However, it turns out that it is worrisome so far for those [ some ] who see.

I think that it will be careful anew and I will see.
Did you understand the difference ?

thanks. ;)

[lunaticmoon]

(About this theme, since a forum participant's abbreviation half was indifferent, I avoided standing a new thread and it decided to divert this thread.)

In the talk which specialized in Anime, the Anime of hybrid material is increasing more recently.

The following two methods can mainly be considered to encode hybrid material smoothly strictly.


A) Interlace maintenance encoding approach:
In the material holding the interlace, it is not careful for especially treating hybrid material. However, the quality of de-interlacer at the time of reproducing this as progressive is asked.

About interlaced encoding, the page of minolta is detailed.
http://www.vrac.iastate.edu/~chadspen/interlace/index.html


B) Progressive VFR (variable frame rate) encoding approach:
The material itself is encoded and saved in the progressive state. In order to treat hybrid material, there is a method of setting to VFR or the method of setting to false VFR (120fps etc.). If the method of thinning out a suitable progressive frame is automatic, it has the problem that processing is difficult.

There is the following about a false VFR (120fps) filter.

1) DEC60() avisynth filter and AVI60.exe
http://www.geocities.co.jp/SiliconValley-Sunnyvale/3109/
http://www.geocities.co.jp/SiliconValley-Sunnyvale/3109/oldtools.html

2) itvfr.auf, vfrout.auo AVIUTL(and avisynth) filter
http://www.tenchi.ne.jp/~yoko/aviutl/vfrout/

3) Auto60.auo AVIUTL filter

No WebPage :(

4) Auto field shift AVIUTL plug-in (AFS)
http://www.geocities.jp/aji_0/


Though it was regrettable, as long as the opinion of people of a forum was heard, by RV10, that it cannot finish treating well made both of the techniques clear.

To an individual, I like the appearance of RV10 at the time of treating Anime. It seems that it becomes impossible however, to solve the present condition of Anime then.

A) won B) in the point of a user's ease of treating.
B) won A) in that it is precisely smooth.

Hybrid material of future Anime tends to increase increasingly. Which do you of this forum choose about this treatment?

Z1) A = Interlace maintenance approach
Z2) B = Progressive VFR encoding approach
Z3) Since trouble is disagreeable, it does not care, even if not smooth.
Z4) I do not treat Anime.
Z5) First of all, it is uninterested.

When it comes to me ... Is it delicate? :confused:

[lunaticmoon]

Automatic field shift plug-in was upgraded to 5.1, while Doom9forum had fallen.

I covered the manual of the attachment over machine translation, and added some proofreading this time.

It seems that different approach from plug-in of other formation of false VFR is used for automatic field shift plug-in.

Now, as for automatic field shift plug-in, the community to which I belong also attracts attention.

Automatic field shift plug-in:
http://www.geocities.jp/aji_0/
Original manual
http://www.geocities.jp/aji_0/afs_guide.html


Automatic field shift interlace release plug-in Guide


1. Introduction

The automatic field shift (afs.auf) is interlace release plug-in for AviUtl. Although operation is checked by AviUtl0.99, 0.98d etc. of reports of operation etc. is obtained also in AviUtl of a former version. This text explains the feature of the automatic field shift, processing, and a function.


2. Feature of Automatic Field Shift

When terrestrial broadcasting is recorded on videotape with PC, the data with which horizontal stripes are contained in the picture with a motion may be created. Such a picture is called interlace picture. The picture containing the horizontal stripes which an interlace (interlace) means that it weaves in by turns, and are obtained on PC videotape recording is the combination of the picture of two sheets.

Terrestrial broadcasting sends the picture which shifted delicately in low resolution every 1/60 seconds. In PC, two sheets are saved combining this picture as a high resolution picture of every 1/30 second one sheet. For this reason, if a difference is in the contents of the combined picture of two sheets, the portion is visible to stripes. In order to distinguish the picture of terrestrial broadcasting, and the picture of PC videotape recording, the picture of 1/30 second unit of the "field" and PC videotape recording is called "frame" for the picture of 1/60 second unit of terrestrial broadcasting.

An automatic field shift shifts the field combination of a frame. (it shifts) It is things and is plug-in which suppresses striped generating. The frame which shifted the field needs to be reproduced to the timing as which the field included in inside was displayed from the first. However, the field cannot reproduce original display timing in the same data form as the video of 1/30 second unit made from PC videotape recording for 1/60 seconds since it was the picture of a unit. If data is saved with output plug-in corresponding to the automatic field shift, video reproducible to original display timing can be created. Moreover, various image processing for removing the horizontal striped pattern which is not lost only in a field shift is carried in the automatic field shift.

Processing of an automatic field shift is close to performing 60fps reading by AviUtl and thinning out automatically one of the frames of even number or odd number. The data created also becomes 60fps accuracy. In 60fps(es)-ization, the image contained in sauce does not need to change processing by 24fps(es) or 30fps(es). An automatic field shift can carry out progressive-izing of the image created for the material of various frame rates in the same processing similarly.


3. Processing of Automatic Field Shift

Two paths are analyzed and a screen is compounded by 3 path eye. (He is what was explained later on and, as for Pass 1-3 of the following, a user does not need to be [ order ] conscious of the processing inside an automatic field shift.)

[Pass1] Field combination is determined.

1) A horizontal striped pattern is detected in each field combination before and behind the present frame.
2) The combination of the field is determined.
3) A result is written in a shared memory so that output plug-in can be referred to.
4) A screen end can be removed from a judgment domain (the width which four directions except is specified with the number of pixels).
5) It is hard to shift the field, so that a shift threshold value is large (it always turns off by 256).
6) Combination of the field is not changed when a field shift is turned OFF.

Setting value related to Pass1: A top, the bottom, the left, the right, a shift threshold value, stripes (shift), field shift

[Pass2] Detailed picture analysis for release

1) A horizontal striped pattern and a motion are detected with the frame of the field combination determined by Pass1.
2) Unevenness of the domain judged to be a striped pattern is accustomed.
3) It moves between the adjoining fields and a judgment is shared.

setting value related to Pass2: stripes (release) and Y -- moving -- C -- moving

[Pass3] Composition of the screen canceled [ interlace ]

1) A screen is compounded by the specified release system.
2) The color for every pixel is calculated with reference to the analysis result of Pass2.
3) Scene change detection (release Lv1) and slanting line compensation (release Lv5) are calculated here.
Setting value related to Pass3: Release Lv, scene change detection (release Lv1)

[Others]

A) Infanticide

Since display timing will become uneven if the field is shifted, it is an input image. A frame with a display period shorter than one frame occurs. Since reproduction of such data is heavy, it is made to the data which is easy to reproduce by thinning out the short frame of a display period. This setup is effective, only when corresponding output plug-in is used, after turning ON the field shift.

B) Smoothing

An automatic field shift determines display timing per field synchronizing with a sauce image. For this reason, the data with which the frame of the sauce 2 field and the frame of the sauce 3 field were intermingled by the display period is created. Although such data is faithful to the display timing of a sauce image, a problem may be in reproduction with PC. In animation reproduction of PC, if control of accuracy cannot be performed for 1/60 seconds or the load of reproduction processing is changed, there are many cases where it becomes impossible to reproduce smoothly. Smoothing equalizes both of display periods in the sauce 2.5 field, when the frame of the display period sauce 2 field and the 3 fields adjoins. The data which can be reproduced comparatively smoothly by this also in the environment which cannot perform exact reproduction is created. This setup is effective, only when corresponding output plug-in is used, after turning ON "a field shift" and "infanticide."

C) Formation of 24fps(es)

At the image which broadcast a movie, animation, etc. on television, it is output frame rate. The data which will be easy to reproduce if it fixes to 1/24 seconds may be able to be created. If this option is turned ON, the display timing of a frame is fixed to one 1.25 times the cycle of sauce. This setup is effective, only when corresponding output plug-in is used, after turning ON "a field shift" and "infanticide." In addition, the error of the �} sauce 2 field arises in the synchronization of an image and a sound by 24fps-ization. It 24-fps(es)-turns partially, and when outputting, the frame just behind 24fps portions may be lost under the influence of with error.

D) The display of a shift judging result

The field combination information on the present frame and the following frame is displayed. When there is a field shift, a red box and the following frame are expressed for the present frame as a blue box. At the time of an output, a display disappears automatically.


4. The example of display timing

example 1)

Sauce 60 field/sec
(Figure)
Automatic field shift Interlace release 30 frame/sec (>X> shifted frame)
(Figure)
Correspondence output plug-in 60 frame/sec (* Null frame)
(Figure)
Infanticide on 60 frame/sec
(Figure)
Smoothing-on 120 frame/sec
(Figure)

example 2)

Sauce 60 field/sec
(Figure)
Interlace release 30 frame/sec
(Figure)
Correspondence output plug-in 60 frame/sec
(Figure)
Infanticide on 60 frame/sec
(Figure)
Smoothing-on 120 frame/sec
(Figure)

The above display timing is calculated by afs_client.h of output plug-in.


5. Striped Release Function of Automatic Field Shift

The image-processing function which removes the horizontal striped pattern which did not disappear only in a field shift is carried in the automatic field shift. A release function can be chosen from 0 to 5 by the release Lv of a setup, and as for Lv0, an afterimage of5 [ release nothing, and / Lv1-Lv5 ] decreases, so that a number is large.

[Release Lv0] With no release

A horizontal striped pattern is not canceled. The new frame together put in the field shift becomes an output as it is.

Setting value related to release Lv0: Nothing

[Release Lv1] Formation of field Mie

The field in front of one more is added to the new frame together put in the field shift as an afterimage, and it is crowded. The striped pattern by motion is completely changed into an afterimage. It becomes a screen near the feeling which watches television with an old cathode-ray tube.

If scene change detection is turned ON, the field in front of one will be added on condition that the following, and it will not be crowded, but will double within a frame.

1) A motion is detected from 75% or more of the all pixel in a judgment domain (it judges with a striped (release) value).
2) The color composition in a judgment domain changes 50% or more.
3) The frame which carried out the field shift does not detect a scene change.

In order to take in the field in front of one, it is the desirable release system which is used in the field shift off.

setting value related to release Lv1: stripes (release) and Y -- moving -- C -- moving -- scene change detection (release Lv1)

[Release Lv2] Striped detection doubleness

A striped pattern is detected within a frame, a striped portion is equalized, and it changes into an afterimage. A movie, animation, etc. have turned to the image with few motions of a field unit.

Setting value related to release Lv2: Stripes (release)

[Release Lv3] Motion detection doubleness

As compared with a front frame, only a portion with the motion (change of a color) equalizes stripes. If it is release Lv2, the stationary horizontal striped pattern equalized can be saved. Please use it to crush the outline of the fine character of the stationary telop, and a still picture with framework, such as animation. Keep in mind that there are side effects, like the omission in release of stripes remains if the setting value of motion detection is raised too much.

setting value related to release Lv3: stripes (release) and Y -- moving -- C -- moving

[Release Lv4] Between motion detection assistant

The portion which had a motion as compared with the front frame is complemented with the picture of the field of the direction which crushes and leaves one of the two's field. As for the portion which required this release, vertical resolution becomes half although an afterimage is lost. It is necessary to set up various kinds of threshold values carefully according to the picture of the object to cancel so that a striped pattern may not be detected in any places other than interchange race stripes to cancel.

setting value related to release Lv4: stripes (release) and Y -- moving -- C -- moving

[[Release Lv5] Slanting line compensation complement

The compensation function of a slanting line is added to the complement of release Lv4. Although slanting edge is complemented smoothly, if a complicated picture is canceled, a noise may occur.

setting value related to release Lv5: stripes (release) and Y -- moving -- C -- moving


6. Updating History

2004/08/07 Automatic field shift ver.5.1 correspondence
2004/08/04 Automatic field shift ver.5.0 correspondence
2004/07/20 Automatic field shift ver.4.6 correspondence
2004/07/19 Automatic field shift ver.4.5a correspondence
Miswritten word correction of a release Lv explanatory note
2004/07/18 Automatic field shift ver.4.5 correspondence
First edition



Please try those interested and be.:)

[lunaticmoon]

In 2nd stage, since it thought that it became the argument by the codec, I moved the thread.

http://forum.doom9.org/showthread.php?s=&threadid=80882