madVR, SVP, and AMD perfect picture and steady vision

[professa X]

Yo

Been ages! 

 

So i got to clarify somethings and im hoping u guys can help.

 

madvr has something called smooth motion. I use it cuz it helps with judder albeit with ghosting. If im correct then madvr use frame blending to achieve this. 

 

Svp is pretty cool too. I havent used it but seems like it makes some smooth video playback. But they are a bit confusing. On the wiki page in tge section "under the  hood" the interpolation describes sounds more like blending, but other google search link titles as well as youtube videos use the moniker frame doubling interpolation. So im confused as to which one it really does. 

 

jRivers wiki implies that frame blending is not interpolation but i think it is. I think interpolation is a general term and madvr uses frame blending while svp uses frame sampling (if its frame doubling) am i right? 

 

I also want to know if AMD's Perfect Picture and Steady Vision is really a one package with stuff similar to madvr and svp in it? 

 

And one final thing. If i have content that is natively 30fps (consistent) and i watch it on a 60hz screen. Would i see a difference vs the same 30fps upscaled to 60fps on a 60hz screen (utilizing either frame blending or frame sampling)? 

 

Would i see a difference between content that is upscaled via frame blending vs frame sampling aside from ghosting that frame blending creates? 

 

Yes i used the terminology frame sampling under the  assumption that frame doubling interpolation that at svp uses is frame sampling. 

 

Please help this driving me nuts! 

 

[Nikolai Smirnof]

There’s a lot of confusion floating about regarding motion smoothing, frame sampling, interpolation, and frame blending. Between jargon in our community and proprietary names muddling the waters, finding common understanding is challenging. Some of these terms refer to the same video processing strategies, but some do not. I’ll also provide links to help people follow my logic.

 

To attempt to make sense of this mess, I’m going to refer to practices used in the animation industry, video processing in Adobe Premiere, and features included in televisions. With those established, I’ll relate those processes to proprietary software like AMD’s Perfect Picture, Microsoft’s Hyperlapse, and Nvidia’s PureVideo. These will also be used as a direct comparison to open source software such as MadVR and Smooth Video Project (SVP).

 

In the realm of animation, there is a terminology known as animating on “1’s, 2’s, or 3’s,” which refers to the number of frames an image is repeated during playback of a video. While 1’s provide more fluidity in video, 3’s can afford more detail. While several framerates are used in video production, such as 24 fps, 30 fps, and even 50 fps, 24 fps is widely held as the industry standard for animation, film, and television. That said, when an animator would say that they are animating in 1’s, each image would have a 1/24^th of a second duration, and so on. This animation standard is well described in this video:

 

https://www.youtube.com/watch?v=5_MDprx3tOU

 

Two video processing methods used in Adobe Premiere are frame sampling and frame blending. Frame sampling has much in common with the 1’s, 2’s, or 3’s principle in animation, in which the duration of existing frames are shortened or lengthened to give the appearance of image smoothness. On the other hand, frame blending (motion interpolation) uses algorithms defined in software to generate new, intermediate images between existing images. Below is an example of 30 fps to 120 fps (4x) motion interpolation:

And a comparison of these video processing methods (optical flow seems to be an artifact-rich process…)

 

https://www.youtube.com/watch?v=DidAd-dyQ2c

 

A caveat of these processes are video artifacts, the severity of which will depend on the software and hardware used for video processing. Below is an example of how poorly an intermediate frame can be rendered:

 

A common feature of modern LCD televisions is referred to as “smooth motion” or “motion smoothing,” which could either refer to frame sampling and frame blending, as defined by Adobe, but your mileage can greatly vary, as there is no industry standard for video smoothing. Other strategies used in the tv industry include black frame insertion and backlight strobing, which attempt to take advantage of the stroboscopic effect to induce image smoothness (poorly). Here’s a good resource on home television shenanigans:

 

https://www.youtube.com/watch?v=KTY9uDD0x14

 

Motion smoothing has been implemented into proprietary software such as AMD’s Perfect Picture and Microsoft’s Hyperlapse, which use motion interpolation in combination with image enhancement features, such as image upscaling and color correction.

 

Similar features are used for motion smoothness and image enhancement in software such as MadVR and SVP. SVP applies motion interpolation using the video processing program, Interframe. These interpolated frames are stored as pre-rendered images into the GPU cache, then are passed to the video player of your choice for rendering (VLC, MPC-HC, MPV, Potplayer, etc.) Image enhancement plugins, such as MadVR or EVR Custom, can then be used to apply further video processing.

 

With all of that established, I’ll try to help point by point.

 

The video smoothing used in SVP is achieved through motion interpolation, which can render videos at a higher framerate (be careful to only do so in whole integer increments to avoid juddering, such as 48 fps, 72 fps, 144 fps for a 24 fps source, and so on). Unfortunately, “frame blending” is a flexible term, but frame doubling likely refers to frame sampling, which is not what SVP does to achieve higher framerates.

 

However, the “smooth motion” feature in MadVR functions using frame sampling, which can be useful when paired with SVP to account for any disparity between the framerate of your source material and your monitor’s refresh rate. This is where whole integer interpolation is critical to avoid screen judder. For example, an interpolated 24 fps video will play without judder on a 144hz refresh rate, but an interpolated 30 fps video will judder at 144hz. A simple check is dividing your monitor’s refresh rate by your source video’s framerate. If you don’t get a whole number, some judder will be present during playback. There is some misinformation about interpolating a 24 fps source to 72 fps on 60hz monitors for smooth playback, but all that will achieve is unnecessary load on your computer.

 

AMD’s Perfect Picture is pretty comparable feature-wise to MadVR and SVP, but I’m unsure of which does it better. Chances are good that choosing the latter will give you much more flexibility.

Not to be picky, but upscaling refers to artificially increasing a video’s resolution, but I think you meant interpolating. Anyway, I definitely think you’ll see a difference if you interpolate 30 fps to 60 fps on a 60hz screen. I’d recommend interpolating and frame sampling, but interpolation will make a much bigger difference.

 

Ghosting is a legitimate concern regarding video interpolation, which frame sampling can somewhat mitigate. However, artifacts such as ghosting, halos, and banding can be reduced mostly by using higher bitrate videos, and using higher refinement settings on your software if your computer has the hardware to support it.

[professa X]

4 hours ago, Nikolai Smirnof said:

There’s a lot of confusion floating about regarding motion smoothing, frame sampling, interpolation, and frame blending. Between jargon in our community and proprietary names muddling the waters, finding common understanding is challenging. Some of these terms refer to the same video processing strategies, but some do not. I’ll also provide links to help people follow my logic.

 

To attempt to make sense of this mess, I’m going to refer to practices used in the animation industry, video processing in Adobe Premiere, and features included in televisions. With those established, I’ll relate those processes to proprietary software like AMD’s Perfect Picture, Microsoft’s Hyperlapse, and Nvidia’s PureVideo. These will also be used as a direct comparison to open source software such as MadVR and Smooth Video Project (SVP).

 

In the realm of animation, there is a terminology known as animating on “1’s, 2’s, or 3’s,” which refers to the number of frames an image is repeated during playback of a video. While 1’s provide more fluidity in video, 3’s can afford more detail. While several framerates are used in video production, such as 24 fps, 30 fps, and even 50 fps, 24 fps is widely held as the industry standard for animation, film, and television. That said, when an animator would say that they are animating in 1’s, each image would have a 1/24^th of a second duration, and so on. This animation standard is well described in this video:

 

https://www.youtube.com/watch?v=5_MDprx3tOU

 

Two video processing methods used in Adobe Premiere are frame sampling and frame blending. Frame sampling has much in common with the 1’s, 2’s, or 3’s principle in animation, in which the duration of existing frames are shortened or lengthened to give the appearance of image smoothness. On the other hand, frame blending (motion interpolation) uses algorithms defined in software to generate new, intermediate images between existing images. Below is an example of 30 fps to 120 fps (4x) motion interpolation:

And a comparison of these video processing methods (optical flow seems to be an artifact-rich process…)

 

https://www.youtube.com/watch?v=DidAd-dyQ2c

 

A caveat of these processes are video artifacts, the severity of which will depend on the software and hardware used for video processing. Below is an example of how poorly an intermediate frame can be rendered:

 

A common feature of modern LCD televisions is referred to as “smooth motion” or “motion smoothing,” which could either refer to frame sampling and frame blending, as defined by Adobe, but your mileage can greatly vary, as there is no industry standard for video smoothing. Other strategies used in the tv industry include black frame insertion and backlight strobing, which attempt to take advantage of the stroboscopic effect to induce image smoothness (poorly). Here’s a good resource on home television shenanigans:

 

https://www.youtube.com/watch?v=KTY9uDD0x14

 

Motion smoothing has been implemented into proprietary software such as AMD’s Perfect Picture and Microsoft’s Hyperlapse, which use motion interpolation in combination with image enhancement features, such as image upscaling and color correction.

 

Similar features are used for motion smoothness and image enhancement in software such as MadVR and SVP. SVP applies motion interpolation using the video processing program, Interframe. These interpolated frames are stored as pre-rendered images into the GPU cache, then are passed to the video player of your choice for rendering (VLC, MPC-HC, MPV, Potplayer, etc.) Image enhancement plugins, such as MadVR or EVR Custom, can then be used to apply further video processing.

 

With all of that established, I’ll try to help point by point.

 

The video smoothing used in SVP is achieved through motion interpolation, which can render videos at a higher framerate (be careful to only do so in whole integer increments to avoid juddering, such as 48 fps, 72 fps, 144 fps for a 24 fps source, and so on). Unfortunately, “frame blending” is a flexible term, but frame doubling likely refers to frame sampling, which is not what SVP does to achieve higher framerates.

 

However, the “smooth motion” feature in MadVR functions using frame sampling, which can be useful when paired with SVP to account for any disparity between the framerate of your source material and your monitor’s refresh rate. This is where whole integer interpolation is critical to avoid screen judder. For example, an interpolated 24 fps video will play without judder on a 144hz refresh rate, but an interpolated 30 fps video will judder at 144hz. A simple check is dividing your monitor’s refresh rate by your source video’s framerate. If you don’t get a whole number, some judder will be present during playback. There is some misinformation about interpolating a 24 fps source to 72 fps on 60hz monitors for smooth playback, but all that will achieve is unnecessary load on your computer.

 

AMD’s Perfect Picture is pretty comparable feature-wise to MadVR and SVP, but I’m unsure of which does it better. Chances are good that choosing the latter will give you much more flexibility.

Not to be picky, but upscaling refers to artificially increasing a video’s resolution, but I think you meant interpolating. Anyway, I definitely think you’ll see a difference if you interpolate 30 fps to 60 fps on a 60hz screen. I’d recommend interpolating and frame sampling, but interpolation will make a much bigger difference.

 

Ghosting is a legitimate concern regarding video interpolation, which frame sampling can somewhat mitigate. However, artifacts such as ghosting, halos, and banding can be reduced mostly by using higher bitrate videos, and using higher refinement settings on your software if your computer has the hardware to support it.

 

Dude u are officially my dude! This is more than i hoped for. Thank you so much. 

 

MadVR: frame sampling. Frames are repeated. 

SVP: motion interpolation. Utilize the frame blending discussed (a new image created via algorithm). 

 

My goal is to watch 24fps animation on a 60hz screen.

 

I have used madvr and reduced judder significantly but u can still see it. I use the setting where it only enables smooth motion if there will be judder so some frames get repeated and some dont.  this helps prevent ghosting vs having smooth motion on all the time.

 

Now it seems the best idea is to use madvrs frame sampling along with svp frame blending. 

 

I feel that  ideally i would want my smooth motion to convert my 24fps to 30fps via frame sampling then take that out put and apply frame blending with a integer multiplication of 2 so i achieve 60fps on 60hz.

 

But i dont know how to achieve this. I feel like i should just convert this stuff first then watch it. 

 

Do you have some pointers for either approach (madvr + svp, convert my self). 

 

Again, thanx! 

[Nikolai Smirnof]

I would recommend against converting a 24 fps source to 30 fps, then interpolate to 60 fps. You will experience amount of screen judder that frame sampling may not be able to cover up. From a few lines of math, the resulting image would be roughly 40% out of synch:

 

Original source ratio: 24 fps / 24 fps = 1

Converted video ratio: 30 fps / 24 fps = 1.25

Amount of desynch: (1-(1/1.25))*2*100 = 40%

 

While I’d generally caution against interpolating to 72 fps, you’ll end up with a smoother playback if you do. While you cannot perceive 72 fps on a 60hz monitor, it will appear smoother than 48 fps or converted to 60 fps.

 

If you want to try 60 fps regardless, there is a feature in SVP for fixed playback at a multiple or framerate of your choice if you want to make a custom profile. Then you can use frame sampling in MadVR to try to compensate for motion judder.

 

It’s very difficult to give advice on MadVR or SVP settings without dabbling with the settings yourself, there’s a lot of trial and error involved. However, when you install SVP, it will scan your hardware and automatically configure optimal settings “under the hood.” SVP has an excellent forum dedicated to helping users, and they just put out a new user manual for SVP 4 which I found pretty useful. For MadVR, the best advice I can give is to look up guides on Kodi, AVS, or other forums that can give a detailed guide on use (it's really dense material). I saved the guides to PDFs, but since then the posts have gone down. If you have any specific questions about MadVR or SVP, I'll share what I know.

Edited September 4, 2017 by Nikolai Smirnof