Speeding up HDTV to DVD via AVISYNTH & VDub

[thymej]

Let first start off by saying sorry for being long winded and I’m no expert by a long shot. I have been taking all the good advice from users on this forum and others and applying them to my en-code processes. This is also posted at Doom9. So here we go:

I thought I would share what I discovered on how I sped up en-coding of HDTV 1080i to DVD 480. It’s a two step processes. There is a big gotcha though; you will need 7 times the disk space of your HDTV source. To compare speed, using the same quality settings for each method, the old way would have taken 12 hours to a 4 hour job with the new way.

Here is how it all started:
I was saddened when I realized I would only average anywhere from 5-8 frames per-second (CCE speed of 0.21) during the en-code. My en-code process was to en-code loading an AVISYNTH script in CCE. So a 1 pass 43min episode would take close to 3hrs to go from HDTV (1980x1080i) to DVD (720x480). With such a slow process I had to settle on a 1 pass or 2 pass en-coding. With a two pass I would have to wait close to 6hrs for just one 43min episode of 1080i. For me, that was to long. What I really wanted was a 4 pass en-code using CCE; but at 5-8 frames per second (CCE speed of 0.21), that’s just too long.

NOTE: My system is a P4 3GHz with 1GB ram Windows XP.

After reading lots of forums threads here and else ware, someone wrote something not related to speeding things up; they wrote telling someone to use VirtualDub and save the video as AVI for testing purposes. So I thought, would CCE take an AVI file from VirtualDub that was saved in a raw video format; raw, meaning no compression whatsoever (VDub’s Direct Stream Copy mode). What I discovered is, not only did CCE load the raw AVI but is also processed it at 15 time the speed. It went from a speed of 0.21 to 3.3. Wow, now I can do a 4 pass and get the quality I wanted.

Pass/Step 1 (3hrs (I need a dual core to make this step faster))
Pre-process the AVISYNTH script through VDub and Save as AVI using the Direct Stream Copy mode. NOTE: this will create an extremely large file. My 5GB 1080i episode of Heroes saves as a 31GB 480p file (Film frame rate IVTCed).

Pass/Step 2 (1hr for a 4 pass CCE, 15mins for a 1 pass)
Process the AVI file from step 1 above directly in CCE. Then muxman your audio.

Here is the script I use in pass/step 1:

MyVideoD2V="C:\DVDTest\Heroes-Test5.d2v"
SetMemoryMax(384)
MPEG2Source(MyVideoD2V, idct=3)
TFM(d2v=MyVideoD2V,slow=2)
TDecimate()
clip1=Spline64Resize(720,480)
masker=greyscale(invert(clip1)).tweak(bright=-90,cont=1.2)
clip1
clip2=HDRAGC(max_sat=1,avg_lum=100,coef_gain=2.5,m ax_gain=14,min_gain=0.5,protect=1,reducer=1)
clip3=overlay(clip1,clip2,opacity=0.3,mask=masker)
clip3
Convolution3D(preset="movieHQ")

It would be interesting to see how much faster it would be for those who en-code SD material.

Then again, the Con to this is it takes lots and lots of disk space; but it works for me. I can live with a 4hr en-code process when I know I’m getting the 4 pass en-code I wanted.

ThymeJ

NOTE: the 4 pass I’m talking about is really a CCE VBR multipass 3 pass. The added 4th pass is CCE in pass1 creating the Video Info File.

[gadgetguy]

I don't remember who or when, but this has been recommended before. On multi-pass encoding it's faster because frameserving from AviSynth processes each frame for each pass. By saving to an interim file, the AviSynth process only happens once. However, you can use a lossless compression like HuffyUV to save disk space. The end result shouldn't be affected.

[thymej]

I actually tried that very thing because I thought the same thing, "that is a large file 30GB plus". I used HuffyUV and saved to avi. The en-code process slowed down to less then half then what it take for the raw video, plus the time it took for en-code/save to the HuffyUV. Since I have the disk space, I decided to stick with raw. Thanks

[manono]

I also save out an intermediate AVI when the filtering is going to be slow, and use Lagarith, as I can save it to the YUY2 color space that CCE prefers. I think it might decode even slower than HuffYUV, though. I think it's HDRAGC that's slowing you down. You might try the AutoLevels filter for AviSynth. If it does what you want, it's one heck of a lot faster, and you might find you no longer need to save out the uncompressed AVI.

http://forum.doom9.org/showthread.php?t=128585

Also, if you don't need the deinterlacer in TFM, you can turn it off and speed up things some more with PP=0. And I'm not sure why you're doing all that filtering anyway, if the source is a good quality Hi-Def video. Like, why C3D at all, or the use of Overlay, or even HDRAGC? Is the source capture that messed up that it needs all that stuff?

[thymej]

The C3D helps remove the flicking noise that was added when the broadcast company comprest the HD source to fit in the OTA bandwidth.

Yes I did notice the HDRAGC slows it down, it slows it down by 2-4 frames per second. The HDRAGC I recently added to bump up the shadows just a very very little. The playback on the PC looks okay but dark on my TV. Even the source is a little dark on my TV. I've tried levels & YLevelsC and they both mess with the color to much. HDRAGC also messes with the color the least. Using overlay with HDRAGC does the best on keeping the color the same as the source.

I will have to look for the autolevels filter for avisynth and try it. You don't happen to have the link?

ThymeJ

[manono]

I gave you the link in my previous post.

[thymej]

Sorry I read right over it. Thanks

[halsboss]

The easy way with minimum fuss :- near the bottom of https://forum.videohelp.com/topic338564.html#1764785
Just adjust from PAL to NTSC settings (576 -> 480 and pal-dvd -> ntsc-dvd) ...

also

https://forum.videohelp.com/topic337233.html#1754361 and
https://forum.videohelp.com/topic337231.html

[manono]

None of that has any relation to this thread. His source is telecined and just needs an IVTC to become progressive again. He's using CCE as his encoder, and not ffmpeg. I'm happy it all works for you, but to me, that doesn't look like The easy way with minimum fuss - far from it.

[halsboss]

Good point, glad you picked it up.

[thymej]

Now I'm looking for a avisynth filter thats faster then hdragc, that brightens the dark areas only without messing up the color to much.

I'll try any and all suggestions.

[manono]

Tweak(Bright=20,Coring=False)#or more, or less

It's built into AviSynth. There's no separate DLL to load.

[thymej]

Cool, Thanks

I did not like tweak when used by itself, it brightens the whole image not just the dark areas; but when used with overlay, I think the Tweak just might work, I just ran a few single frame tests. I will now have to run a longer test clip through.

I changed the tweak to Tweak(Bright=30,Coring=False) since the overlay is set low (30%). I also tweaked the mask clip to mask more of the brighter parts of the image out. I also changed the overlay mode to 'Add' so the darker areas show through brighter with less tweak setting. Here is my new overlay section:


masker=greyscale(invert(clip1)).tweak(bright=-120,cont=1.25)
clip1
clip2=Tweak(Bright=30,Coring=False)
clip3=overlay(clip1,clip2,Mode="Add",opacity=0.3,m ask=masker)