What 10-bit colorspace can I use with avisynth?

[-Habanero-]

I only found ConverttoYV411 so far and I installed that modded FFMS2.DLL but virtualdub refuses to open it because it needs the Y41B codec which I can't find any information from.

[newpball]

You might be confused by the term 10 bit colorspace and chroma subsampling, a commonly used method to cut corners with color.

[jagabo]

It appears to be a 4:1:1 planar layout. If you can't find a decoder you can probably change the fourcc and the image dimensions, then use some creative cropping/stacking to get everything in the right place. Can you upload a short sample?

A hex dump will make it obvious if the data is 8 bit or 10 bit.

[-Habanero-]

Oh wow, YV411 is not a 10-bit colorspace? This microsoft site says Y410 is 10-bit and 4:4:4 so I confused the two.

My clips are 8-bit, they are all different recordings of the same video which I wanna mix together to create an HQ version. But I'm getting banding which pisses me off so I wanna convert them all to 10-bit before mixing them this time.

I followed http://avisynth.nl/index.php/High_bit-depth_Support_with_Avisynth#Importing_High_Bit-d..._into_AviSynth but now I need a 10-bit format to convert to.

[jagabo]

x264 supports 10 bit encoding.

[-Habanero-]

First I need it in 10-bit format, I get banding if I mix them in 8-bit.

[newpball]

First I need it in 10-bit format, I get banding if I mix them in 8-bit.

You are mixing different recordings of the the same video in the believe you will get high quality.

How do you suppose that could possibly work?

[-Habanero-]

http://www.digitalfaq.com/forum/video-restore/2734-averaging-multiple-captures.html
It already is a lot better quality, completely devoid of noise or other momentary artifacts like dropouts. DCT blocking is also gone.
Try and be a little less close-minded.

Either way, I didn't start this thread to discuss the merits of my method but to ask how to convert to a 10-bit format.

[newpball]

DCT blocking is also gone.

That implies you are averaging lossy compressed captures.

I am open minded but I don't think much good can come from that.

By the way we are talking about analog captures right?

[-Habanero-]

That implies you are averaging lossy compressed captures.

Yes, and some have signal where others have loss. Mixing them together gets rid of all temporal defects. The only way for defects to survive is if they all simultaneously have the same defect in the same spot.

Either way, this is off-topic and not what I started the thread for. Can Avisynth work with 10-bit video or not?

[newpball]

That implies you are averaging lossy compressed captures.

Yes, and some have signal where others have loss. Mixing them together gets rid of all temporal defects.

Okay...if you say so.

Either way, this is off-topic and not what I started the thread for. Can Avisynth work with 10-bit video or not?

First it would help if you re clear what you want. Do you want to tinker with the chroma subsampling or do you want deep color or perhaps both?

Your problem is you have 8bit video, you can't make that into 10bit.

I really do not want to discourage you but your idea of averaging a bunch of lossy compressed videos is not going to magically make things any better or 10bit. The only thing that averaging can do is to statistically remove "transmission" (the signal read from tape up to the capture moment) errors, noise or errors already on the analog carrier will no go away.

[jagabo]

Your problem is you have 8bit video, you can't make that into 10bit.

Yes, he can. He's averaging together multiple videos. For example (1+0+0+0)/4 = 1/4. That's two more bits than the individual sources.

[newpball]

Your problem is you have 8bit video, you can't make that into 10bit.

Yes, he can. He's averaging together multiple videos. For example (1+0+0+0)/4 = 1/4. That's two more bits than the individual sources.

Explain that to me, so you have an 8bit video recording and you capture that many times lossy compressed and then eventually it turns out 10 bits? Is that your theory?

[poisondeathray]

Mathematically, jagabo is correct.

Each pixel in each channel can have 4 possibilities between each "run" , and can have a value between 0 to 255 if we are talking full range values. So this means each pixel can have a value of 0 to 4x255 possiblities, which is 10bit

However, that says nothing about the quality of the 10bit

Can Avisynth work with 10-bit video or not?

That's not really your question, is it ? ie. you don't have a 10bit source

Eitherway, avisynth is 8bit natively only. Vapoursynth can work natively at higher bit depths

But there are workarounds to work in 16bit stacked format in avisynth using dither tools. Read the documentation in the dither tools - it gives you examples of how to do this to an 8bit source and send it to 10bit x264 for example

To answer the question you actually asked, you can load 10bit sources with the special 10bit ffms2 "hack" version which returns the 16bit stacked version

[-Habanero-]

Some screenshots to shut the ignorant newbball up: http://screenshotcomparison.com/comparison/122532

That's not the video I'm currently working with and banding isn't particularly visible there.
And just so you know, the first screenshot is from the best recording out of the 10. The others are worse. Yet notice how the averaged photo looks so much better than even the best individual recording in the pack.

Explain that to me, so you have an 8bit video recording and you capture that many times lossy compressed and then eventually it turns out 10 bits? Is that your theory?

Are you familiar with 1-bit photos? 0 or 1, black and white. Merging two together within the same bitdepth is useless because the average of a 0 and a 1 pixel would be 0.5 which would be rounded to 1. In other words, they'll always mix to black.

But convert those photos to 8-bit greyscale and black+white pixels will become gray. You just went from a photo with only two colors to a photo with three different colors.

Now mix many different 1-bit photos of the same shot and you might get a high quality greyscale 8-bit image.

Is this elementary logic clear to you now? Maybe if I didn't lose the password to my last account, you wouldn't feel so free to confidently preach to someone who's been here 6 years longer than you.

[newpball]

Are you familiar with 1-bit photos? 0 or 1, black and white. Merging two together within the same bitdepth is useless because the average of a 0 and a 1 pixel would be 0.5 which would be rounded to 1. In other words, they'll always mix to black.

But convert those photos to 8-bit greyscale and black+white pixels will become gray. You just went from a photo with only two colors to a photo with three different colors.

Now mix many different 1-bit photos of the same shot and you might get a high quality greyscale 8-bit image.

Fascinating.......



It is a sad thing though, nobody apparently wants to tell you that averaging repeated captures will not remove compression artifacts and neither increase the color depth of the video.

[-Habanero-]

Fascinating, DURRRRRRP!

I just had to test it out after mentioning it. Here's a 1-bit video and the composite average. 256 frames of all the possible exposures from brightest to darkest.
Evidently, 256 1-bit photos have the same entropy as one 8-bit photo. Just like 4 8-bit videos have the same entropy as one 10-bit.

Listen more, preach less and you might not have to roll your eyes so much in the future.

[poisondeathray]

It is a sad thing though, nobody apparently wants to tell you that averaging repeated captures will not remove compression artifacts and neither increase the color depth of the video.

His example shows how averaging can reduce noise (increased signal to noise ratio), but it's not a demonstration of 8bit vs 10bit from multiple captures.

Averaging (as in "mean average", or "averaging" captures the way it's typically done) won't get you higher bit depth (you don't sum and then divide by 4, otherwise you're back to square one, ie. 0-255 values) - you need to keep the 4 sums in 10bit and this becomes almost true 10bit (there is a bit of error in the calculation, you have a bit fewer values than true 10bit)

But you can increase the bit depth, if you do it correctly. You can generate a 10bit video from 4 * 8bit values - this is a fact , at least mathematically .

And in practice, you DO get more than 0-255 valid possible values. And you do get smoother gradients.

Sort of related , but this is what people do to get 10bit Y' at 1080p from UHD 4:2:0. Because the UHD 8bit Y' signal is essentially 4 "boxes" of 1920x1080, you have the same math 4x0 to 4x255 possibilities . It works, and you can verify this in something like after effects to read the pixel values . 8bit has quantized "jumps", but the downscaled 1080p has smooth 10bit values , at least in the Y' channel. (You would need 4:4:4 at UHD to get 10bit CbCr at 1080 - ie. there is only 1 "box" at 4:2:0 because of the subsampling)

In real life, typical footage, it's not that simple - there is actually a bit more to it . There is the effect of dithering for depth conversions (this actually has more of an impact in real life). Also, you are usually limited by compression artifacts , quantization, and low sensor latitude in the first place on typical footage

I can show you some proof on some 8bit test gradients if you want, or give you instructions on how to "see" 0-1023 values in AE

[-Habanero-]

That's not really your question, is it ? ie. you don't have a 10bit source

Eitherway, avisynth is 8bit natively only. Vapoursynth can work natively at higher bit depths

But there are workarounds to work in 16bit stacked format in avisynth using dither tools. Read the documentation in the dither tools - it gives you examples of how to do this to an 8bit source and send it to 10bit x264 for example

To answer the question you actually asked, you can load 10bit sources with the special 10bit ffms2 "hack" version which returns the 16bit stacked version

Vapoursynth, I'll have to check that one out. I already have the special ffms2.dll so I probably could load a 10-bit clip if I had one.
What I need to know is how to upscale my current clips to 10-bit, but even then I'm not sure if the avisynth functions I'm using would work in that bit-depth. I'm currently using Overlay() to average the clips. If that won't work in 10-bit then I'm screwed and will have to invest in a professional video editor in the future.

[poisondeathray]

Overlay() works in 8bit . But you should be able to use the stack16 version to do the operation with dither tools - this should keep the LSB values

If your source captures don't have "banding", then you suddenly have "banding" - then something is wrong with what you are doing .

Either you are denoising too much (noise acts as a natural dither), or you're not using enough bitrate / high quantizers (ie. "banding" from macroblock edges) . Does the "banding" occur before you encode it? (look at a script preview). And what type of "banding" is it ?

[-Habanero-]

Overlay() works in 8bit . But you should be able to use the stack16 version to do the operation with dither tools - this should keep the LSB values

How? Dither_convert_8_to_16() and then do everything else like I normally would? The documentation isn't clear on this.

If your source captures don't have "banding", then you suddenly have "banding" - then something is wrong with what you are doing .

My source captures don't have banding. Here, I mixed all the frames of this one 45-minute episode into one frame using Overlay. Notice that even this gets banding: