iip: Integrated Image Processor, e.g. DVD -> 720p HDTV

[Didée]

>> How "simplified" are we talking here?

Oh, I was thinking about temporalsoften(5,255,255...) for denoising, 3 or 4 times blur(1.58) for deringing, and unfilter(150,150) for fine sharpening

Seriously. SSXsharpen is ultra-slow cause of the big supersampling, and will be replaced by LimiteSharpen as default (like in the alpha), and optionally by the even faster ModerateSharpen. This also helps the warp_Y option, in case it gets used.
PixieDust most probably will get LRemoveDust as faster option.
For deringing I've also another way of creating the mask, which is both a little faster and better fitting.

[DarkNite]

Sounds like a solid plan. These are the kind of script changes that can make comparison testing fun again.

As an added bonus, another 100 or so variables means more fun when randomly generating "the perfect settings" when someone asks for help with a source.

Seriously, that was just a joke. Do not do this to your freinds on irc, unless you're at least an op, or your nick contains the word Evil.

[guada 2]

Hello everyone,

To this address: http://forum.doom9.org/showthread.ph...&pagenumber=10).
I tried this script for the encoding of the movie matrix3 in 1024*576 with the RV10 format, and hfe 2.1 activated.
The result is without call. I also used the Xvid codec but the result was not the same.
But, when I wanted to test NIC'S Wmv9 program: WMNicEnc v.0.99 bêta, the script doesn't function.

Would someone be able to explain me why?.

[dark.soft]

Hi Didée, I tried your script and I must say that is outstanding! But I have two questions for you:
1\On things that aren't on focus (e.g. walls), the script applies the sharpening and they appear ugly, like if they are pencil coloured! What are the parameters wich I should change to correct this?
2\I tried a certain piece of Resident Evil, the elevator scene, where there is a lot of blue. Well, with your script I noticed a thing that happens only when converting to yv12: the borders of the blue are blocky, and if I set a resolution like 640x480 they become really noticeable. So, can you tell me a solution?
Thank you very much for writing such an amazing script!

[guada 2]

Sorry, Didée. It works very well with WMNicEnc v.0.99. i just missing an dll in Windows.
Your script is fantastic.

[PlazzTT]

There's not much mention of using VHS sources in this thread, although on other threads it has been mentioned that iip is great for processing VHS.

Which settings should be adjusted from default, taking into account that a noisy VHS source is being processed, and not a relatively cleaner DVD, TV or HDTV source?

[dark.soft]

I've made other tests, and I've noticed that all the things that are blurry (not on focus) are sharpened bad, so that they seem in focus, but that's wrong. I don't think there is a solution for this, am I right Didée?

[Didée]

guada 2: Fine if it works and pleases.

dark.soft: If blurry background objects are enhanced too strongly, then

- set protect_floor to zero if it isn't already
- raise protect_bias
- raise detail_floor
- drop the "EQ" parameter by 1, and reduce the sharpening settings
- (less probably) certain fiddling with deringing can help, too

Use the debug="showall" feature for tweaking the parameters:

- The "protection" mask shows you which parts of the frame are protected from getting sharpened, and how strongly.

- The "detail" mask shows you in which places the sharpened result is applied. If there are too much visible "structures" in those areas that are getting too much sharpening, raise "detail_floor" until these disappear. [edit]


PlazzTT: I've very little personal experience with processing VHS captures. First it must be assured that the souce is sufficiently "calmed down" by the internal denoiser. Raising "duststr" to 5 or 8 at max. might be necessary, together with both deflicker options @ "true". If this doesn't suffice, you need to use another denoiser beforehand.
Then VHS may need more deringing - or it may not. Depends on your signal.
For sharpening, you probably should raise radius to 3 or even 4, especially when [stronger) deringing is used. But when using radii > 2, detail_contr1 should be lowered somewhat, mostly.
Perhaps raising raising detail_contr2 and/or pixsharp is beneficial, if the source is clean & stable enough for that.
Warp_UV could be a good idea for VHS, too.

Much of poking around. I don't do any VHS, and input quality might be just anything, so it's hard to judge.

[guada 2]

Hello Didée,

I tried your script on an extract of matrix 2 with a bitrate of 450 Kbps in 720p.
I have been very surprised of the quality of the picture.

My file of 23.5mo has been divided by 10 either 2.35 mo.
Even in pause on fast stages, the script preserves its chromatic wealth.

I wonder what your script would give on a cd of 700mo in HD.
To be honest, it is one of my main objectives.

[Didée]

Oh, indeed. HD @ 450 kbps.

\/

[dark.soft]

720p at 450 kbps? Guada 2, you're a genius!

[Soulhunter]

Ouch, now I understand why he gets better results with RV10...


Bye

[guada 2]

Hello Gentlemen,

One would say that you didn't appreciate my bitrate.
I think that I had forget to specify something,it is HD streaming in the WMV9 format and no of the RV10 (Mr SOUL my Friend).

I saw it was possible to put on a SD card of 512 mo 2 Xvid Movies.
So I do some test in Streaming HD 720p on 800mo.

Think yourselves that I am mad or brilliant

[JasonFly]

I'm wondering what is the "weakest link" of the processing chain with such low speed scripts?
Is it CPU, RAM, HHD speed?

BTW, what if this script would become a plugin? I don't know if it would be possible but could there be a speed increase by making that?

[DarkNite]

In my experience this particular script will bring your cpu to it's knees. iip could replace prime95 as your burn in test.

Memory usage can be "controlled" (ie "MemoryMax(192)") within the avisynth script, assuming there's no leaks in the functions/plugins you're using, or apps you're serving to, or anything else involved in the chain. So using 192MB-256MB of memory isn't bad at all considering what you get out of it.

I can't speak for Didée, but making this a plugin would limit some of the flexibility this script offers to people who choose to alter the script to meet their needs, or the needs of their source.

Who knows, you might see somebody do a plugin of the "simplified", less slow version that's been mentioned. Stranger things have happened here.

Lastly, HDD speed is always a problem in computing. Lobby for faster development of nonvolatile dynamic random access memories (like FDRAM).

[DeepDVD]

As a silent protest I used my freetime to make some banners for some demos to get the new iip. See my signature

[Didée]

Nice try, DeepDVD

But you have to know I only fall for 3D renders. If possible, with radiated illumination.

This was done 5 years ago, when I wondered about possibilities for a PC desk in that darn corner in the living room with rustical furniture:



(POV-Ray/MegaPOV, 100% scripted. No pre-made libraries or textures were used, except painting & screenshot.)


To ease the task for you: radiation isn't mandatory. But 3D is.

[guada 2]

Hello Didée,

In pre-visualization before the validation of your script, Nero recode read perfectly my file, but after to have validated the file, I note that it is reversed in mode reading.
Normal or not normal ?

[akupenguin]

Originally posted by JasonFly
I'm wondering what is the "weakest link" of the processing chain with such low speed scripts?
Is it CPU, RAM, HHD speed?

I'm guessing RAM. It uses lots of simple filters, which do only a little processing each but still require generation of a new frame. And many of them apply to a high-res image that won't even fit in L3 cache. I'm sure it could be made several times faster by processing in slices.
However, I agree that packaging it in a plugin would greatly reduce the easy of tweaking it.

HDD speed only matters for very fast filterchains, and only if one or both ends are huffy.

[JnZ]

Hi,

I'm just testing IIP, and my results are fantastic. Didée, you are genius.

Script is very slow, about 1-2fps on my Athlon in 720p res., but with incredible result!

So, if we really want convert DVD -> 720p, it's better:
1)make one pass to HuffYUV, and from this source make 2-pass XviD encoding (finaly 3-pass).
2)or make directly 2-pass XviD encoding, but both pass with very slow speed (1-2fps).
This is the really question!

I think, that first possibility is better, because hardest work (IIP resize) can be shared to two or more computers. Video can be divided to many parts and result in HuffYUV can be joined. Anyway in 1), IIP script can be anytime stopped, and next time we make encoding on next frame, and finally join this avis. Etc.

Thx. Didée, now, I'm thinking, that hollywood makes HI-DEF from normal DVDs, but uses your IIP!