不知道怎麼吐槽了，AMD 用的技術叫 Convolution-based interpolation 而不是你說的 frame blending，madVR 的 smooth motion 才是基於 frame blending 的概念



Fluid Motion Video，透過AMD專有的Frame Rate轉換技術，能夠在進行24fps轉換成60fps的視訊播放時，降低畫面換格時的接續不順，並提供更流暢、更平順的播放畫面。
一般的24fps轉60fps的視訊轉換時，原先是1-2-3的播放順序，轉換成1-1-1-2-2-3的播放順序，亦即前三張的畫格其實都是複製第一張的內容，這種作法在播放時會有Video Judder(晃動)的狀況。(高速移動最明顯)
而AMD的Fluid Motion Video則是參考第一張和第二張中間的變化，安插新的變動畫格，因此產生出1-1a-1b-2-2a-3的播放結果，讓視訊在任何Frame Rate下都能夠平順、不損畫質地播放。

This mode of operation is based on the idea to treat the time dimension of a video clip as just another static dimension, the same as the resolution - essentially, we are treating the input as a big 3D signal of size Width×Height×Duration, reconstructing this input using standard techniques from signal theory. Basically, instead of convolving the kernel with discrete pixels in the source image, we convolve the kernel with the same pixels across discrete frames.

The main difference to smoothmotion is that it essentially low-passes the time axis to make sure no high frequency distortions get added (which can look like irregularities with the smoothmotion algorithm), and that it can reconstruct some “intermediate” pixel values, which works well for slow motion in particular, but can add extra motion blur in some cases.

As you can see, the overall appearance looks similar to smoothmotion, but instead of frame transitions alternating between sharp and blended, each frame is always blended into the next one.

你說的 frame blending

The smoothmotion approach is to display each frame exactly 2.5 times, where we display frames for 0.5 times by blending two adjacent frames together. In terms of the pattern, it results in something like this

A A A+B B B C C C+D D D E E E+F F F

Here, A+B refers to a frame that is exactly half of A and half of B, blended together. The overall result is that each frame is displayed for a consistent amount of time, resulting in smooth motion.

As you can see, the overall appearance is smoother due to the increased regularity of the frame heights, and the overall perceived framerate (24 Hz) has not been affected - the lines are still clearly distinct.

This is not only easy to calculate (no motion-based prediction whatsoever, just a simple blending operation of two frames), but also preserves the perceived framerate of the original clip, so there's no soap opera effect or similar.

Essentially, smoothmotion is like using a nearest neighbour resize but oversampling the result (similar to how multi-sampling works in video games to reduce aliasing). Due to this, it's called "oversample" in the mpv implementation.

你覺得好棒棒的 SVP

Motion-based interpolation algorithms like SVP, MVTools or the chips built into various TV devices employ complex algorithms to try and recognize movement in the video (motion vectors), and fill in the missing gaps as needed.

This drastically alters the result - it's now one continuous motion, similar to a true 60 Hz clip. However, this comes at a cost - not only are the mathematics involved extremely expensive to compute, but the algorithms also aren't perfect, and often result in artifacts (deformed images, wavy lines, etc.).

Furthermore, this is the source of the so-called soap opera effect, which is simply referring to what 60 Hz motion looks like. The name is based on the fact that cheap TV programs like soap operas are often filmed at interlaced 60 Hz rather than the (more expensive) film format that most cinematic content uses, which is usually shot at 24 Hz. Many people therefore subconsciously associate the visual appearance of 60 Hz content with soap operas, and thus think it looks worse.

This is essentially similar to "smart" upscaling filters, eg. NEDI or NNEDI3 - which are also expensive to compute and produce results that drastically alter the source image.

無論是 interpolation 還是 frame blending 都不是原始影片中的產物，為啥 frame blending 在你眼****來的畫面是錯誤的；interpolation 就是正確的我實在想不明白

（更遑論你根本搞錯概念....）

嚴格來講 SVP 出來的 interpolation 也是視覺暫留作用（你要說腦補也行），阿我是不是要說請格放出來找出問題點？

madVR 的 smooth motion 基於 frame blending，難道出來的畫面也是錯的？這基準到底誰說了算？你嗎？

我的看法79樓表達得很清楚了，極度要求 smooth 且不 care artifacts，那當然 SVP 比較好

你下一篇就回 BFRC 有補跟沒補根本一樣，浪費100多瓦的電；SVP 就是補的好但錯誤少，看了不想吐槽都不行

先不論其背後的技術原理到底為何，至少 BFRC 的效果是明顯可見的，而且用肉眼觀察也看不出有什麼不自然之處；而 SVP 的 artifacts 明顯可見