25fps PAL 576i ➔ 50fps 1080p.
DV AVI ➔ AviSynth QTGMC + FineDehalo ➔ AI image resize.

Upscaling a video with AI image resizers. For example, interlaced SD DV video to HD at 50 or 60 fps. This could still be improved a lot by utilizing temporal information and motion flow. And the speed isn't yet too impressive. For now, processing a single frame takes 1...5 seconds, depending on your HW (mine's pretty old...).

Updated version of the soundtrack: [ Ссылка ]

The videos in this example were recorded with Canon XM2 video camera.
My AviSynth scripts processing the original DV AVIs (before resizing) looked like this:

vid1 = "some_DVCam_video_file.avi"
SetMTMode(5, 4)
vid1 = DirectShowSource(vid1, audio=TRUE)
vid1 = ConvertToYV12(vid1)
SetMTMode(2)
vid1 = QTGMC(vid1,
\ FPSDivisor=1,
\ TR0=0, TR1=0, TR2=0,
\ Rep0=0, Rep1=0, Rep2=0,
\ NNeurons=1,
\ NNSize=2,
\ Lossless=2,
\ Sharpness=0,
\ ShutterBlur=0,
\ ShowSettings=FALSE)
Distributor(vid1)
vid1 = FineDehalo(vid1, rx=3.0, ry=2.0, darkstr=.5, brightstr=.75)
# vid1 = Trim(vid1, 110, 310) # to pick a desired part of the clip
return vid1

# improvement suggestions appreciated :-)

свернуть