madVR Options Explained

[Arm3nian]

Also check this: http://i.imgur.com/7VaN0Tc.png

Notice how SSAA is referred in both the horizontal and vertical resolution. 2*2=4, so 2x2SSAA is shortened as 4xSSAA. This can be confirmed by looking below at the combined SSAA and MSAA options. 32xS is 2x2SSAA + 8xMSAA, 2*2=4, 4*8=32. This is the most logical method of labeling the values.

[Asmodian]

Replace resolution with 'information'. If something has 4x the information of another thing, saying it has 2x instead makes no sense. If you want to say 2x the resolution, then you would have to independantly define and state that resolution x is 2 times greater horizontally and 2 times greater vertically than resolution y.

This is exactly my point. The sort hand "2x the resolution" means 2 times greater horizontally and 2 times greater vertically. You cannot replace resolution with information; they are not synonymous.

To be precise "2x the resolution" is ambiguous, it could have either interpretation. It would be more precise to say "2x the resolution horizontally and vertically" or "twice the number of pixels".

There are reasonable reasons to talk about resolution as the total number of pixels but that is not the convention normally used when discussing digital images and video (unless you are a marketing department).

2x scaling gives 2x the resolution (Nvidia's DSR notwithstanding, they only want to describe the performance hit and do not care about conventions).

Anyway I am happy to agree to disagree, only know if I say "2x the resolution" I mean 4 times the number of pixels.

[Arm3nian]

This is exactly my point. The sort hand "2x the resolution" means 2 times greater horizontally and 2 times greater vertically. You cannot replace resolution with information; they are not synonymous.

To be precise "2x the resolution" is ambiguous, it could have either interpretation. It would be more precise to say "2x the resolution horizontally and vertically" or "twice the number of pixels".

There are reasonable reasons to talk about resolution as the total number of pixels but that is not the convention normally used when discussing digital images and video (unless you are a marketing department).

2x scaling gives 2x the resolution (Nvidia's DSR notwithstanding, they only want to describe the performance hit and do not care about conventions).

Anyway I am happy to agree to disagree, only know if I say "2x the resolution" I mean 4 times the number of pixels.

I'm just going off the fact that the term resolution was created to describe information. You can say that 3840x2160 is 2x the resolution of 1920x1080, but what does that mean? It is a useless definition if resolution does not imply information. Maybe it does make sense when strictly talking about photography or video, but that is because we already know doubling the resolution in both directions leads to 4x the pixels, and are just trying to talk about the size. Resolution on its own is a meaningless value if not talking about information. In engineering, the resolution bandwidth of a spectrum analyzer for example implies more samples, therefore shows more information.

[Asmodian]

I'm just going off the fact that the term resolution was created to describe information. You can say that 3840x2160 is 2x the resolution of 1920x1080, but what does that mean?

It means there are twice as many pixels, both horizontally and vertically; you can resolve details that are half the size. :p

It is a useless definition if resolution does not imply information. Maybe it does make sense when strictly talking about photography or video, but that is because we already know doubling the resolution in both directions leads to 4x the pixels, and are just trying to talk about the size. Resolution on its own is a meaningless value if not talking about information. In engineering, the resolution bandwidth of a spectrum analyzer for example implies more samples, therefore shows more information.

When analyzing information the purpose of using the term "resolution" instead of "samples" is to signify the relative spacing of samples in a data set. In a 1D set twice the resolution is twice the samples, in a 2D set twice the resolution is 4 times the samples, in a 3D set it is 8 times the samples, 4D 16, 5D 32, etc. If you want to discuss the total number of samples instead use "samples", "pixels", or similar.

If you have twice the resolution you expect to be able to resolve details at half the size. For a 2D space you need four times the number of pixels to resolve details at half the size. e.g. If the smallest filament you can resolve with an imaging system is 0.1 mm in diameter and you need to be able to resolve one with a diameter of 0.05 mm you need a sensor with double the resolution. It is not useful to raise the change in resolution required to the power of the number of possible axes when discussing a needed change in sampling. It simply removes meaning. Saying you need 8 times the resolution to resolve a 3D object at half the size is pointless, why not say you need 8 times the samples if that is how you want to express it?

Resolution is a term used to describe data sets which are both a type of information themselves and contain information. The relative change in resolution equals the relative change in the number of samples only when talking about a data set with one dimension.

At this point I think I have expressed my option on the matter as well as I can. :o

[Arm3nian]

It means there are twice as many pixels, both horizontally and vertically; you can resolve details that are half the size. :p



When analyzing information the purpose of using the term "resolution" instead of "samples" is to signify the relative spacing of samples in a data set. In a 1D set twice the resolution is twice the samples, in a 2D set twice the resolution is 4 times the samples, in a 3D set it is 8 times the samples, 4D 16, 5D 32, etc. If you want to discuss the total number of samples instead use "samples", "pixels", or similar.

If you have twice the resolution you expect to be able to resolve details at half the size. For a 2D space you need four times the number of pixels to resolve details at half the size. e.g. If the smallest filament you can resolve with an imaging system is 0.1 mm in diameter and you need to be able to resolve one with a diameter of 0.05 mm you need a sensor with double the resolution. It is not useful to raise the change in resolution required to the power of the number of possible axes when discussing a needed change in sampling. It simply removes meaning. Saying you need 8 times the resolution to resolve a 3D object at half the size is pointless, why not say you need 8 times the samples if that is how you want to express it?

Resolution is a term used to describe data sets which are both a type of information themselves and contain information. The relative change in resolution equals the relative change in the number of samples only when talking about a data set with one dimension.

At this point I think I have expressed my option on the matter as well as I can. :o

I agree on all points. I think the problem we have is that 'resolution' on its own is a very ambiguous term. Are we talking about image resolution, pixel resolution, spatial resolution, or even resolution that has nothing to do with digital pixels like my spectrum analyzer example.

Nvidia most likely uses 4x in the DSR to show performance impacts, as the gpu will render 4x the amount of pixels. For photography and video I see why 2x would make sense. 4x when talking about scaling might give the wrong impretion on expected quality. The size of the image is important to consider. It would be to the benefit of all if everyone was more descriptive, but as you already know, it is hard to get an entire industry to agree on something.

[newguy1]

Isn't the whole 2x or 4x resolution an industry standard that the 4k tv decided to overlook?

You know, a standard? Like from a textbook?

[Warner306]

Isn't the whole 2x or 4x resolution an industry standard that the 4k tv decided to overlook?

You know, a standard? Like from a textbook?

The marketing always seems to skew towards selling more televisions as opposed to providing technical information, but I don't know why this simple topic has to cover so many consecutive posts.

A 720p source has 2.25x the number of pixels of a 1080p source. However, the actual increase in vertical resolution or (pixels per inch) is only 1.5x. The amount of dots per inch is a better indication of quality than the total number of pixels.

Moving from 1080p to 4K, the number of pixels increases to 4x, but the scaling factor only improves to a mere 2x.

A better question is whether anyone can resolve the extra pixels per inch from their seating distance, which they probably cannot.

[Arm3nian]

The marketing always seems to skew towards selling more televisions as opposed to providing technical information, but I don't know why this simple topic has to cover so many consecutive posts.

A 720p source has 2.25x the number of pixels of a 1080p source. However, the actual increase in vertical resolution or (pixels per inch) is only 1.5x. The amount of dots per inch is a better indication of quality than the total number of pixels.

Moving from 1080p to 4K, the number of pixels increases to 4x, but the scaling factor only improves to a mere 2x.

A better question is whether anyone can resolve the extra pixels per inch from their seating distance, which they probably cannot.

Everyone in this thread understands 4x vs 2x. We were just debating on the correct terminology. In my first post I mentioned 1920x1080 is 4x the resolution of 3840x2160. Saying 2x horizontal and 2x vertical resolution is the same. 2x scaling just makes it confusing. 3840x2160 is 4x more demanding than 1920x1080, so we should refer to it as 4x unless otherwise specified that you are talking about the pixel count in 1 direction.

Think about a water bottle. If I have a water bottle that is 1ft tall, and I increase it to 2ft, I now have a water bottle that is 2x bigger, because it holds 2x the water. If I increase it to 4ft, then I have a watter bottle that holds 4x as much as water as the initial bottle, making it 4x bigger. Now lets say I invent a gun that makes objects 2x bigger and I shoot my water bottle. Is it going to make it only 2x taller? No, it is going to multiply ALL dimensions by a factor of 2. Now the water bottle does not hold 2x as much water, it holds 8x as much water. So in theory, increasing something by a factor of 2 is relative on the amount of dimensions. In 1d, a factor of 2 increases the size by 2, in 2d, a factor of 2 increases the size by 4, and in 3d, a factor of 2 increases the size by 8. In other words, it more precise to refer to these things by what matters. In resolution, the pixel count, which directly relates to the performance. And in the case of the water bottle, to the volume, which directly reflects the amount of water it can hold. Making something 2x bigger just implies that the actual value did not increase linearly, so why not just refer to the actual value directly... makes it more simple, and more logical.

Resolution is just another word for pixel count. A resolution of 2 means 2 pixels. A resolution of 3840x2160 means 8294400 pixels. The image is just the object. The resolution is the quantity being defined. Just like a water bottle is the object and the volume is being defined. This is why madshi refers to it as image doubling, because making an image 2x bigger creates 4x the pixels. Saying the resolution in 1920x1080 to 3840x2160 doubled makes no sense, because it quadrupled. The image itself doubled.

4k might provide an increase in perceived image quality on a TV, but it is more noticeable on a monitor to me since I sit close.

[resides]

Thanks for the write up.

Definitely helped me and clarified things too.

[Asmodian]

you could add IVTC to deinterlacing. and you my add an comment that force film mode doesn't work with native DXVA

I never use DXVA decoding but I feel that adding notes on its ramifications where appropriate could be helpful. Are there any other details or issues you know of with DXVA native decoding? Thanks.

Thanks for the write up.

Definitely helped me and clarified things too.

Thanks, I am glad it helped a bit. :)

[huhn]

I never use DXVA decoding but I feel that adding notes on its ramifications where appropriate could be helpful. Are there any other details or issues you know of with DXVA native decoding? Thanks.

nothing limited to madVR. just the general. impossible to use the xy/vsfilter, some renderer can't display subtitle. nothing that should be part of this in my eyes. just basic stuff like this.

maybe this:
madVR can't accept p010 from DXVA native. i'm not even sure if there is a stable software that is trying to do this.

[MSL_DK]

image up or downscaling isn't used when displaying 1920x1080 content on a 1080p display, correct? I'm asking because i've read that upscaling is used on everything, perhaps i've misunderstood that article.

[Qaq]

Its chroma upsampling that is used always, cause 4:2:0 needs to be upsampled to 4:4:4.
1080@1080 doesn't need any image upscaling, but keep in mind that some *1080* videos are not exactly 1080 (1077 or so).

[Asmodian]

I did some minor updates and corrections and I expanded the calibration section some. yCMS deserves a mention at least.

maybe this:
madVR can't accept p010 from DXVA native. i'm not even sure if there is a stable software that is trying to do this.

This is good to know but I am not sure where to put it. Maybe simply as a note near the end?

[sheppaul]

I did some minor updates and corrections and I expanded the calibration section some. yCMS deserves a mention at least.



This is good to know but I am not sure where to put it. Maybe simply as a note near the end?

What about creating the section like "known problems" or "limitations"?

[huhn]

This is good to know but I am not sure where to put it. Maybe simply as a note near the end?

this is only affecting the gtx 960 at the moment no other card can hardware decode 10 bit sources at least AFAIK.

and it was not 100% correct from me to say it can accept it. it simply shows a green/black screen when native DXVA feeding P010 in madVR.

some other limitations i know are:
madVR IVTC can detect a lot of cadences but can "only" handle a couple of them correctly.
it can handle at least 3:2, 5:5 and 8:7 from a 60i source. very rare cadences like 3:2:3:2:2 (25 fps in 30i 60 fields) 4:2:2:2 doesn't work

madVR currently always switches to 23Hz when having forced film mode on while playing 59i content. Yeah, with some cadences 23Hz is not the optimal refresh rate. But that's not so easy to solve, especially because cadence detection can be unstable. E.g. what happens if cadence detection switches between 3:2 and between 3:2:3:2:2 all the time? Should madVR then always switch between 23Hz and 25Hz all the time, in the middle of playback? This is a problem I'll have to revisit later, but I don't consider it a "bug" right now. It's as intended, even it might not be the best (or even the correct) solution.
source: http://bugs.madshi.net/view.php?id=150

the decimation feature is currently not working the way it should be. see here: http://bugs.madshi.net/view.php?id=199&PHPSESSID=af06a17857165747bc51e1f417cb10e6

madVR can't do frame packed 3D

and this from the madVR thread
known problems / limitations:

- rotated displays (portrait vs landscape orientation) don't work yet

[michkrol]

First of all, great guide. It must have taken a lot of work to write such extensive explanations.

You've asked for suggestions, so here are mine:
File name tags:
madVR will look at the file name for "magic words" or tokens that tell it to treat a particular file differently.
Technical nuance: madVR looks for tags in the whole path, so both folder(s) and file names.
Pro Tip: put tags in folder(s) name(s) if you have a whole collection of broken rips (files), to save a lot of work (time).
You can even put tags in folders' names on different level for maximum flexibility, for example to get a monster like this:
D:\Movies\TV Rips levels=tv\Broken PAL Rips matrix=PAL\my broken rip from the 90s.mkv
Possible file name tags include:
(...)
What you've listed is of course correct, but you're missing a few tags and possible values.
Tags with missing values or missing altogether (all taken directly from changelog.txt included with madVR):

* added tag "matrix=2020|709|601|NTSC|PAL|YCgCo|240M"
* added tag "primaries=2020|DCI|709|SmpteC|EBU|sRGB|NTSC|PAL|470M|240M|170M"
* added tag "levels=PC|TV|fullrange|limited|doubleExp|tripleExp"
* added tag "blacklevel=%value%", value range [-50, +50]
* added tag "whitelevel=%value%", value range [-50, +50]
* added tag "contrast=%value%", value range [-100, +100]
* added tag "brightness=%value%", value range [-100, +100]
* added tag "saturation=%value%", value range [-100, +100]
* added tag "hue=%value%", value range [-180, +180]
* added tag "frameRate=%value%", e.g. 23.976, 24.000, 23, 24, ...
As a side note you can use either range=... or levels=... tag for the same effect.

Thanks for the guide and keep up the great work :thanks:

[Asmodian]

What about creating the section like "known problems" or "limitations"?

I agree, we have gathered enough items for this section. :)

this is only affecting the gtx 960 at the moment no other card can hardware decode 10 bit sources at least AFAIK.

and it was not 100% correct from me to say it can accept it. it simply shows a green/black screen when native DXVA feeding P010 in madVR.

some other limitations i know are:
madVR IVTC can detect a lot of cadences but can "only" handle a couple of them correctly.
it can handle at least 3:2, 5:5 and 8:7 from a 60i source. very rare cadences like 3:2:3:2:2 (25 fps in 30i 60 fields) 4:2:2:2 doesn't work


source: http://bugs.madshi.net/view.php?id=150

the decimation feature is currently not working the way it should be. see here: http://bugs.madshi.net/view.php?id=199&PHPSESSID=af06a17857165747bc51e1f417cb10e6

madVR can't do frame packed 3D

Good info, Thanks!
These kinds of things could take a long time to debug on your own. :p

First of all, great guide. It must have taken a lot of work to write such extensive explanations.

You've asked for suggestions, so here are mine:

Technical nuance: madVR looks for tags in the whole path, so both folder(s) and file names.
Pro Tip: put tags in folder(s) name(s) if you have a whole collection of broken rips (files), to save a lot of work (time).
You can even put tags in folders' names on different level for maximum flexibility, for example to get a monster like this:
D:\Movies\TV Rips levels=tv\Broken PAL Rips matrix=PAL\my broken rip from the 90s.mkv

What you've listed is of course correct, but you're missing a few tags and possible values.
Tags with missing values or missing altogether (all taken directly from changelog.txt included with madVR):

As a side note you can use either range=... or levels=... tag for the same effect.

Thanks for the guide and keep up the great work :thanks:

Great info! Thanks a lot. Good pro tip, I hadn't thought of using them in folder names. :o

[huhn]

here is my source for the DXVA native P010 problem:
http://forum.doom9.org/showpost.php?p=1710491&postcount=305
it is a green screen. i guess this information is crucial because i can't reproduce this problem on my own. i don't have a hardware decoder for 10 bit HEVC to output p010.

and it looks like you skipped the decimation feature from madVR created for 720p60/720p50 broadcast with 24p/25p film content.
madVR IVTC can decimate progressive streams. just press control + alt + shift+ t to force it on. it has the same limitation like madVR IVTC. in the end it's the same program working.

here the part out of the change log. it is one of the newest feature.
* added support for decimating 50p/60p movies to 25p/24p

[Asmodian]

here is my source for the DXVA native P010 problem:
http://forum.doom9.org/showpost.php?p=1710491&postcount=305
it is a green screen. i guess this information is crucial because i can't reproduce this problem on my own. i don't have a hardware decoder for 10 bit HEVC to output p010.

and it looks like you skipped the decimation feature from madVR created for 720p60/720p50 broadcast with 24p/25p film content.
madVR IVTC can decimate progressive streams. just press control + alt + shift+ t to force it on. it has the same limitation like madVR IVTC. in the end it's the same program working.

here the part out of the change log. it is one of the newest feature.
* added support for decimating 50p/60p movies to 25p/24p

Thanks again, I had missed that decimation feature. I seem to have hit the limit for characters per post so I was unable to add the section. I may edit for reduced character count. :(

Other Features:
-madVR IVTC can decimate progressive streams such as 720p60/720p50 broadcast with 24p/25p content. Press control + alt + shift+ t to force it on. It has the same limitations as madVR IVTC.

edit: some minor edits for length allowed me to add the Other Features section. :)

[Asmodian]

Updated for madVR v0.87.15

Some big fixes. Portrait mode displays (rotated) work! Also Optimus black screen fixes.

There is a bug when picking the color matrix so BT.601 is selected for HD resolution video in some situations. Video flagged correctly is handled correctly as far as I have tested. Most likely a new build soon.

edit: fixed, up to madVR v0.87.17 now.

[Asmodian]

Updated to v0.87.21

A 64-bit build is now available. :)

No major changes in the options.

[Karl78]

Great job Asmodian. Thanks!

[Magik Mark]

Hey Asmodian,

Maybe we can update the guide for v881? Thanks for the awesome work

[huhn]

i recommend to wait for some time X-)

[Asmodian]

Yes, I hope and expect the options change a lot in the near future but I will still attempt an update for v0.88.2 soon. However, it will not cover the fine tuning options for the new enhancement shaders or similar details.

[Asmodian]

Updated for madVR 0.88.4, as always please offer corrections and suggestions. :)

D3D11 support allowing 10-bit output and a collection of sharpeners. Another major update!

[tobindac]

"present a frame for every VSync: when disabled madVR only presents new frames when needed, relying on D3D11 to repeat frames for each VSync."

Search the last week's pages of the madvr thread about this. He said that we should disable it for more speed if we have certain setups. I don't recall how but I'm in that group, I have an R9 290 on Windows 8.1 64.

[Asmodian]

Yes, are you suggesting I add a note that it might help performance? I don't seem to have a change in performance with it off or on.

[tobindac]

I doubt it's going to be noticeable to most. I guess it's some kind of minor improvement of the kind of 0.1-0.2% or something. He said the option solves problems in some setups but in others it's only some latency.

[Warner306]

If Upscaling Refinement shaders are added to the post-resized image, why do they only impact videos that are sub 1080p?

[Asmodian]

If Upscaling Refinement shaders are added to the post-resized image, why do they only impact videos that are sub 1080p?

I assume you are watching on a 1920x1080 display? They apply for me watching 1080p on a 2560x1440 screen. Think of them as part of upscaling and upscaling doesn't run when viewing 1080p at 1920x1080.

[Warner306]

Ok, one more question. Is SuperRes a form of image doubling, or can it do irregular upscales?

[Shiandow]

It can handle arbitrary upscales.

[Asmodian]

I understand SuperRes is more of a post process method than an independent scaler; it runs along with or on top of the other scaling option chosen, it doesn't replace it.

[Shiandow]

I understand SuperRes is more of a post process method than an independent scaler; it runs along with or on top of the other scaling option chosen, it doesn't replace it.

That's more or less correct, although it will still give reasonable results even if you use a really simple scaler such as nearest neighbour, or even if you just feed it a black image. But if you start with a better image you need less iterations, so it's a bit of a trade off.

[nevcairiel]

Personally I'm still not sure about SuperRes. On high quality content, the difference it makes is very minimal, and on low quality content it tends to sharpen the artifacts more than I would like.

[Asmodian]

I do think I like SuperRes for chroma. The minimal difference on high quality content seems like a good difference as well, at least with animated content.

[leeperry]

SuperRes: This is a post process method. From Shiandow (http://forum.doom9.org/showthread.php?p=1685124#post1685124)
Sweeet, thanks for digging the link for us as I was really looking forward to explanations regarding this feature because to my eyes NEDI+SuperRes can really match/exceed NNEDI3's PQ but I'm still finetuning settings atm :)

[Shiandow]

There have been some changes to the algorithm though, although the general idea has remained more or less the same. The steps now look more like the following:


Calculate an initial guess (using a different upscaler)
Downscale and calculate differences with original image.
Scale those differences to the final size
Improve guess by:

Softening the image
Subtracting differences with the original image
Sharpening (optional)
Removing aliasing (optional)
Removing ringing (optional)

Repeat steps 2-4 several times.

[Warner306]

I do think I like SuperRes for chroma. The minimal difference on high quality content seems like a good difference as well, at least with animated content.

How can you tell it's impacting the chroma? I thought it only applied to the luma, or the luma and chroma together.

[huhn]

How can you tell it's impacting the chroma? I thought it only applied to the luma, or the luma and chroma together.

because superres can be used for chroma only.

[leeperry]

SuperRes is chroma only?

Anyway, major kudos to Shiandow once again for the NEDI+SuperRes combo because to my eyes it doesn't have the artificial look of NNEDI3, doesn't look oversharpened and looks very natural/realistic AFAICS. motionDSP's fluff (http://www.motiondsp.com/products/IkenaForensic) doesn't look all that impressive anymore.

Would be fun to know how the combo fares with the samples in this link (http://www.infognition.com/articles/video_resize_shootout.html) but some of those SuperRes apps are far from realtime so this might be quite unfair.

[Warner306]

SuperRes is chroma only?

Anyway, major kudos to Shiandow once again for the NEDI+SuperRes combo because to my eyes it doesn't have the artificial look of NNEDI3, doesn't look oversharpened and looks very natural/realistic AFAICS. motionDSP's fluff (http://www.motiondsp.com/products/IkenaForensic) doesn't look all that impressive anymore.

Would be fun to know how the combo fares with the samples in this link (http://www.infognition.com/articles/video_resize_shootout.html) but some of those SuperRes apps are far from realtime so this might be quite unfair.

Can someone explain the three different SuperRes options?

[Warner306]

There have been some changes to the algorithm though, although the general idea has remained more or less the same. The steps now look more like the following:


Calculate an initial guess (using a different upscaler)
Downscale and calculate differences with original image.
Scale those differences to the final size
Improve guess by:

Softening the image
Subtracting differences with the original image
Sharpening (optional)
Removing aliasing (optional)
Removing ringing (optional)

Repeat steps 2-4 several times.


This should be added verbatim to the guide.

[huhn]

SuperRes is chroma only?

you can use it as the chroma resizer too and don't have to resize luma at the same time. so it can be used for chroma only.

i guess this is more accurate.

[Asmodian]

SuperRes is chroma only?

SuperRes can be chroma only (check the "activate SuperRes filter" box on the chroma upscaling page).

This should be added verbatim to the guide.

Yes, I really wish I had more characters to use so I could. :(

What sections should/could be removed or pared back?
edit: I squeezed it in, I hope I didn't remove anything important. ;)

There have been some changes to the algorithm though, although the general idea has remained more or less the same. The steps now look more like the following:


Calculate an initial guess (using a different upscaler)
Downscale and calculate differences with original image.
Scale those differences to the final size
Improve guess by:

Softening the image
Subtracting differences with the original image
Sharpening (optional)
Removing aliasing (optional)
Removing ringing (optional)

Repeat steps 2-4 several times.


Thanks for the excellent and succinct explanation of SuperRes, I am sure many people with appreciate it. :)

...and all the work you put into the method itself, of course.

:thanks:

[Warner306]

Can you read the following post? I think your rendering diagram is wrong. In the post below, madshi explains the input for image doubling is "always RGB." This is not reflected in your diagram: http://forum.doom9.org/showpost.php?p=1721413&postcount=29697.

[Asmodian]

But my diagram does have the input for image doubling as always RGB? The "image doubling" madshi is talking about is the dashed orange box.

I leave out that madVR will use any of the Bicubic opitons if you have it set for image upscaling but I assume most users will not have a bicubic set for image upscaling. Otherwise that post agrees with my diagram.

[Warner306]

But my diagram does have the input for image doubling as always RGB? The "image doubling" madshi is talking about is the dashed orange box.

I leave out that madVR will use any of the Bicubic opitons if you have it set for image upscaling but I assume most users will not have a bicubic set for image upscaling. Otherwise that post agrees with my diagram.

So you are saying a color space conversion from RGB to YCbCr is a necessary part of image doubling? This is what I questioned based on that post.