|
|
IVTC/Resizing filter order with HDTV (possible speed up?)
I'm bringing back some HDTV (1920x1080) captures to normal D1 proportions and I was wondering if resizing horizontally before doing IVTC would give a possible speed-up with no quality loss.
Normally I'd use this script:Code:
MPEG2Source().Crop(0,0,0,-8)
Telecide(guide=1).Decimate()
Lanczos4resize(720,480)
I wondered if I could replace it with this:Code:
MPEG2Source().Crop(0,0,0,-8)
Lanczos4resize(720,1080)
Telecide(guide=1)
Lanczos4resize(720,480)
Decimate()
I understand I will have to measure the speed-up (if any) myself, but I was just wondering if there would be any pitfalls I am not aware off that might influence quality. If I understand correctly resizing horizontally should be safe for the combing, but I could be wrong.
Thanks for any answers and insights.
[edit]corrected script syntax[/edit]
That should be OK, except for your giving wrong syntax for Lanczos4Resize().
It should give a decent speedup.
Ah yes, let me correct that before anyone sees it.
I made did a small speed comparison using the 'time'-function from MSYS together with avs2yuv, on a small (5000 frames after decimation) sample. This is the command I used in MSYS for it:
Code:
time avs2yuv script.avs -o NUL
(I used NUL output to avoid writing a big raw file to the harddisk)
The old/normal script rendered at 6.8 fps and the new/alternative one at 9.3 fps, giving a 37% speed-up. Very welcome for my not-so-fast machine.
I also found out about a function called 'compare' in Avisynth, which I used to compare the output of the two scripts. It gave me an overall PSNR of almost 60 dB, so I guess the difference is negligible.
All in all, very nice to see how a little reordering of the filter-chain can help so much.
Originally Posted by GodofaGapAll in all, very nice to see how a little reordering of the filter-chain can help so much.
For sure. Thanks for pointing it out. |
|