Once again: HDV soft when rendered to DVD

johnmeyer wrote on 5/5/2008, 3:47 PM
I know this has been discussed, but I still can't get really soft results when rendering HDV from my FX1 to MPEG-2 from which I make a 4:3 NTSC DVD.

1. I render using Best.

2. In the project properties I have a deinterlace method selected (I've tried both Blend and Interpolate -- even though there shouldn't be any deinterlacing taking place since my source is interlaced and I am creating a DVD that is interlaced for display on an NTSC monitor.

I crop the HDV source to 4:3 on the timeline so I don't get black bars.

Any ideas or advice from someone that has been able to get sharp as a tack results from their HDV source, please let me know. I just did a three-camera shoot, with two DV and one HDV camera, and the HDV footage looks worse than the other two.

I'm using Vegas 7.0d.
[Edit] P.S. One setting in the Render As dialog that may have a bearing on this and has always been a mystery is the "Allow field-based motion compensation" setting. This is what it says in the Vegas help file:

Select this check box if you want to use field- and frame-based motion prediction when predicting frames. When the check box is cleared, only frame-based motion prediction is used.

Now this is darned curious because the default in all the templates is for this to be checked, not cleared!!


Comments

winrockpost wrote on 5/5/2008, 4:01 PM
maybe because you are zooming also to create 4:3 .,, , i have had good luck with hdv to widescreen dvd, looks very sharp to me, but never made a 4:3 from hdv . have no idea about Allow field-based motion compensation.
DJPadre wrote on 5/5/2008, 6:34 PM
its HD theres no zooming. all hes doing is shaving off the sides

umm... where are u seeing this motion compensation thingy? coz i sure cant find any reference to this in vegas itself...
johnmeyer wrote on 5/5/2008, 7:13 PM
where are u seeing this motion compensation thingy? coz i sure cant find any reference to this in vegas itself...

It's found if you click the Custom button in the Render As dialog, after you have selected MPEG-2 as the render format. The dialog is shown below:

farss wrote on 5/5/2008, 8:15 PM
I've never played with that setting and I seriously doubt it has anything to do with the problem at hand.

Basically your problem is the old sharpness V resolution problem.
When you downscale you downscale the amount of Edge Enhancement (EE) as well as the resolution. SD tends to use more EE than HD (yucky). SO if you want your downsampled HD to look as sharp as regular SD then you can do one of two things:

1) Wind up the EE in your HD camera.
2) Add EE after the downsample.

1) In my opinion is a really bad thing to do because:

a) If you ever want to use that footage as HD then that EE is cooked in.
b) You might have to try to undo it to avoid line twitter anyway when you downsample.

If you use 2) then make certain the EE is added AFTER the downsampling. I use the Unsharpen Mask FX in Vegas.

All of the above depends on the camera you use and what mode you shoot in. The EX1 in progressive after downsampling to SD can deliver too much resolution leading to line twitter problems. You might need to use GB or Median FXs to reduce res prior to downsampling.

Your other cancern about having to de-interlace is fairly easy to explain. You can't easily downscale a field. What happens with Vegas is it seems to use the nearest line and horrid interlace aliasing is the outcome. I guess in theory what should happen is each field is interpolated in some manner to a full frame, downsampled and then a field extracted.

Bob.
johnmeyer wrote on 5/5/2008, 9:32 PM
Bob,

Very interesting post. Thank you! I have read other posts that suggested adding sharpening, but until I read your post, I figured that was just a band-aid that didn't really address the underlying problem. However, if I understand what you're saying, if I want to match the video from my SD cams when doing a NTSC DVD, then it would make perfect sense to add the EE, as you call it.

I realize that you are down-under in PAL (or is it LAP down there?) land, so settings may be different, but do you have any starter settings to suggest? I have not used sharpening on any project, so I'm not familiar with that plugin.
farss wrote on 5/6/2008, 4:57 AM
It's really a matter of taste. I just add enough Unsharpen Mask until I can just notice it kicking into the fine detail while monitoring on a CRT. If there's any noise left after the downsample I don't use any though.

Bob.

Edit: Glenn did suggest using the convolution kernel but it's a bit of a beast I've yet to tame.
NickHope wrote on 5/6/2008, 6:17 AM
John if you still have trouble after those suggestions, maybe you could give TMPGenc Plus 2.5 a try. In my testing way back it was sharper than Vegas' MainConcept encoder. You can frameserve to it.

It's $37 but there is a fully functional trial that works for 14 days.
johnmeyer wrote on 5/6/2008, 9:35 AM
maybe you could give TMPGenc Plus 2.5 a try.

I still have the TMPGEnc shortcut on my desktop, but only use it to check field order and a few other tricks. I cut my teeth using it to encode to VCD, SVCD, and eventually some of my early DVDs, but haven't used it for years. If I get a chance later this week, perhaps I'll do some test encodes with Vegas, and if I do, I'll frameserve to TMPGEnc as well and see if that looks better.

Thanks!
farss wrote on 5/6/2008, 2:51 PM
One thought though John.
Even rendering from HD to SD DV you'll see the difference you're seeing going straight to mpeg-2. Trying different encoders adds another layer of complication into the equation.
In other words, get the downconvert as good as possible and then find the best encoder to go from there. If TMPGenc does deliver a superior result compared to MC then you should be using it for everything. End result would be that native SD would still look better than downconverted HDV!
Serena wrote on 5/6/2008, 9:38 PM
This has been discussed a few times but while noting the advice I haven't encountered problems that made me get to grips with it. Speaking only of my FX1/Z1 sourced output, mostly this has been produced on DVD widescreen (PAL) and has been good. Usually I use Cineform DIs, so I don't know if this helped. However on one job the producer insisted on staying with m2t (Z1) though post and the definition on the DVDs was fine. I've yet to put 1920 x 1080P to DVD and I have Bob's advice printed and in the tech folder!
johnmeyer wrote on 5/7/2008, 12:39 PM
OK, I did some tests. The results are interesting, but unfortunately not conclusive.

I did the tests using fashion show footage (stage lighting, black backdrop). I rendered a reference test clip from a selection on the Vegas timeline using the Vegas MPEG-2 encoder using the standard 4:3 NTSC DVD Architect template, but changed to 7,000,000 bps average render bitrate. That clip took 0:37 to render.

I then added the unsharp mask fX with settings of 0.2, 0.005, 0 and rendered. Render time increased to 1:28.

Then the same thing with settings of 0.2, 0.020, 0 (the "light" default), and rendered. Same render time as before, 1:28.

I then used these same settings and changed the pre/post toggle for the unsharp mask fX to pre (arrow pointing left). Render time skyrocketed to 5:09. Obviously is takes a lot of time to apply this fX directly to the HDV video prior to downconverting to the project resolution (which was 4:3 NTSC 720x480, BFF).

I then removed the fX to get back to my starting point, but this time rendered with the "No Field Base" setting unchecked in the Render As dialog.

Next, I re-checked that setting to get back to my reference configuration, but this time frameserved, using RGB24, directly into TMPGEnc. I started with the DVD template, but changed from constant quality to constant bitrate, and set that bitrate to 7,000,000. Also, I changed Motion Search to High Quality (slow).

Finally, I frameserved into the standalone MainConcept MPEG-2 encoder, and set the bitrate to 7,000,000 constant, and reduced the noise reduction all the way.

These frameserved renders took only 0:24 for TMPGEnc and 0:20 for the Mainconcept encoder.

Results

In the end, I have no idea.

If I had to pick the one thing that definitely looked "sharper" without looking artificial, it would be the 0.2, 0.005, 0 settings for the unsharp mask. The default "light" settings were way too much. However, if there was ever anything that defined personal taste, this is sure a great test for that. If you put a dozen people in a room, I don't think you'd get anything approaching a consensus.

The pre- post- versions of the same light sharpening settings seemed indistinguishable to me. Obviously no need to spend almost five times the rendering time.

TMPGEnc looked a little washed out, which made it impossible to compare.
If I had a little more time, I'd go back and re-render with the "Output YUV data as basic YbcCr not CCIR601" setting checked (it was un-checked).

I would also like to spend more time fooling around with the "Allow Field Based Motion Compensation" setting in the Vegas MPEG-2 advanced settings. I think it might be making a small difference, but the way I tested was to put the results on a DVD, and then play that onto a large screen Pioneer CRT-based monitor, followed by a similar test on a 30" Sony Wega CRT. However, the tests simply played one after the other. I could go back to the DVD menu and immediately skip to any given test, but I was relying on my memory of what I just saw to make the comparison. Somehow I need to be able to put two MPEG-2 files side-by-side, without any further resizing, in order to really tell the difference. I think that would require two identically calibrated monitors, running next to each other, and then sync up the video from two identical DVDs, each playing a different clip.

As it is, I am afraid that I don't have any conclusions, other than the fact that using the unsharp mask with extremely low settings can make the video appear sharper, but at the expense of double the render time.