Originally Posted by assassin
Eyes can be tricked.
The placebo effect is very real. If there is truly a difference you should see it in the screenshots. Even the Madvr creator uses zoomed in screenshots as his "proof" over at doom9.
While I agree that screen grabs should show differences in "static" IQ such as color space decoding, resolution, contrast, etc, much of the difference I see attributed to various renderers is motion-related. This will not be detectable from isolated screen grabs of single frames.
The part that puzzles me is why a render engine would even affect the "smoothness" of motion (a description I often see applied to madVR) outside of any framerate translation done in the renderer. If a 23.976 fps video is rendered and sent to a 23.976 fps display, I fail to see where apparent motion can be affected since the integrity of the frames should be unchanged. All the render engine should affect is static IQ, which should be visible in screenshots. In the cases where madVR is handling framerate conversions, then all bets are off.
I would expect there to be 4 main factors that influence the perceived video quality - the decoding of whatever codec is used (MPEG 4, VC-1, etc), any framerate translation performed (e.g. 3-2 pulldown), deinterlacing of 480i or 1080i material, and the rendering of the uncompressed frames into raster images to be played back by whatever playback software is used. This of course assumes the playback software is error free, which may not be the case (framerate synchronization or colorspace management being the normal suspects). Two of these are not in play for the playback of progressive material at its native framerate.
BTW, examination of the comparative screenshots reposted in this thread yields extremely subtle differences to my eye, so small as to be insignificant at any real-world viewing difference.
I should add that I am a newbie at HTPC (and Microsoft's Directshow architecture), but experienced in imaging software on other platforms, so I am trying to get a handle on this not from empirical experience with this particular software, but from a more abstract, architectural perspective.
What am I missing here?