Best way to downconvert HDV?
 

Is there any quality difference between these two methods of downconverting?
1) Downconvert right out of the camera/deck and capture in DV, then dupe onto DVD.
2) Capture in HDV and downconvert from the editing system to DVD.

With the FX1, you can literally set the firewire to output to HDV or DV. Set it to DV in the menu and capture in DV mode onto your NLE.

The HD10 allows you to do a downconvert (or upconvert) by using video and audio cables to another deck or camera. You can go 1080i, 480i or 480p (or no conversion).

heath

There is a significant quality difference.

Rendering from HDV (1440x1080 4:2:0 15-GOP MPEG-2) to DVD (720x480 4:2:0 MPEG-2) will yield much better results than rendering to DV (720x480 4:1:1 DV) and then to DVD. Not only does the DV render introduce extra artifacts at the 720x480 level associated with DV compression, but it also requires two colorspace conversions (unless you're working with PAL - in which case DV is also 4:2:0).

Here's the trick though, converting HDV to uncompressed SD for a DVD-render, or exporting from HDV directly to DVD is non-trivial.

-Steve

When you say converting HDV to uncompressed SD, do you mean via analog component cables?

No - I simply mean capturing the MPEG-2, and rendering it to an uncompressed file on down-conversion - retaining as much color information and resolution as possible in the process.

I see no benefit to capturing an "uncompressed SD signal" out of the analog outputs of the camera after recording HDV to tape. I don't know if the camera outputs uncompressed on SD component out before DV compression on down-convert - but in principle, capturing the MPEG-2 TS does not degrade the digital source on tape, while capturing the analog out will introduce conversion errors and noise.

Capturing uncompressed SD output live increases the operating cost, the necessary amount of hardware, and decreases the portability of the camera.

My suggested workflow for 1080i (to maximize quality) is:
- Shoot 1080i HDV (to tape)
- Capture 1080i HDV to M2T or a (virtually lossless) digital intermediate (i.e., Cineform)
- Place 1080i in a 1440x1080, 59.94 fps timeline, and deinterlace (upper fields first, highest quality available) to 1080p60
- Render 1080p60 to 720x480p60 uncompressed (4:4:4)
- Render 720x480p60 to 720x480i60 uncompressed (4:4:4)
- Compress uncompressed 720x480i60 to DVD MPEG-2.

If working with 1080p30 (i.e., Cineframe30 mode)
- Shoot 1080p30 HDV
- Capture 1080p30 HDV to M2T
- Place in 1440x1080 29.97 fps timeline
- Render to 720x480p30 uncompressed
- Compress to progressive DVD

In my experience, these operations can be precisely done in After Effects. Getting fields to behave appropriately in Premiere Pro doesn't seem to work as well.

-Steve

Cineform is not lossless.

Obviously not. But for this purpose, and most, if not all purposes, save perhaps advanced keying, it's really dang close.

I've edited my post to clarify.

-Steve

I'm confused--you want to do uncompressed?

heath

>>> you want to do uncompressed?

Er, no ... he wants to make a DVD.

Simpler Workflow??
 

Steven
What about the following workflow for SD DVD:
Capture m2t from camera (4:2:0), convert to Cineform CFHD (4:2:2), edit the project in CFHD, render the timeline to 720x480 m2v (4:2:0), author the DVD. What would be the disadvantage of this pathway??
Bob

There's no real difference between this work flow and the one I suggested above in an ideal world. However, my experience so far has been that Premiere Pro does not handle this step very well - it has the tendency to bugger the fields or simply ignore them.

I just deconstructed SD renders to ensure proper handling of the fields, before doing the render to m2v.

-Steve

Steven
Does the field order need the be reversed in the process of going from HDV (m2t) to m2v DVD (I'm thinking HDV is upper first--is DVD mpeg lower first, like NTSC DV)?? Or is it just something buggy about Premiere's ability to transcode CFHD to m2v aside from field order issues?
Thanks
Bob

Alternative Route
 

If Premiere cannot reliably convert CFHD to m2v DVD, I was considering another possible route:
Capture m2t into Premiere, convert and edit in CFHD, recompress the timeline back to m2t-as if you were going to export back to HDV tape- then use other software such as Canopus Procoder 2 to convert the m2t to m2v DVD. Has anyone done anything like this??
Bob

Thats what I do. Edit everything in CFHD, then as a final step convert to SD DVD using Procoder. I add a bit of temporal smoothing filter and gamma adjustment during that compression - the advantage of that workflow being that I can store the unfiltered master for future use.

Works fine.

When I rendered the 480p60 to 480i60 I made the field order lower field first just to be sure. I didn't do an experiment with upper field first, so I don't know if it would playback properly or not.

-Steve

Field Order
 

Graham
When you convert CFHD to m2v with Procoder, do you set Procoder for "lower field first" for the m2v?
Bob

What? PPro can't properly write CFHD to m2v? It has to be third party or requires an interim file (i.e. avi and let encore do the transcode?)

Pete

Peter
The problem, as I understand it, with going CFHD>avi>m2v is that you downsample that nice CFHD 4:2:2 color space to DV 4:1:1, and then resample again for DVD to 4:2:0. Ideally I'm looking for a workflow that preserves as much of the HiDef HDV color sampling as possible. The object is to produce SD DVDs that are better looking than a project shot on standard DV.
Bob

I was refering specifically to PPro not be able to properly render a CFHD file straight to m2v. (Which Steven seems to be indicating). This would effect creating DVD's straight off the time for preview, or having to create and intermediate file that can transcoded by a third party ap.

Steven indicates that such is a limitation of PPro.

Do I understand this to be correct?

Pete

>>When you convert CFHD to m2v with Procoder, do you set Procoder for "lower field first" for the m2v

Robert - yes. I usually start with one of Procoder's bundled DVD presets and tweak it. All those presets (whether for NTSC or PAL) specify lower field first.

FX1 HDV->DV into Premiere
 

I am shooting in HDV on an FX1 capturing to Premiere in DV (I'm waiting about a year before I upgrade to a computer capable of the HDV workflow).

Premiere's Media Encoder Presets are all defaulted to: Field Order=Lower. The footage I have been shooting and rendering to .m2v has been coming out as a little pixelated around the edges of objects when I pan or shoot moving objects, BUT I was using the Preset that was 1 pass 7mb CBR High Quality (it said high quality, and 7mb was the highest number, must be the way to go, right? uhh right??).

Having just learned how to create a custom preset (I reccommend Adobe Premiere Studio Techniques Book and accompaning DVD) I created one that was 16:9 NTSC VBR High Quality 2 pass, Field Order=Upper. There was an immediate quality improvement, major, no pixelation. However, there is one section in the video that has a rather quick pan, during this section the whole screen has a rapid side to side movement creating a blurry effect (as if you took a copy of the footage, moved it a 1/2 inch over, and strobed them both through each other).

The book made me aware of Field Order and the difference between VBR and CBR. Was my improvement strictly due to switching to VBR 2 pass and was my jittery side to side motion on 1 of over 15 shots caused by switching to upper field order? I switched the encoding preset field order to upper after reading HDV had an upper field dominance. However, I can't find any info that states if that dominance is preserved or reversed when you downconvert from HDV to DV from the camera into Premiere. I am currently rendering the same file with a lower field dominance with my fingers crossed, but I wanted to bounce it off more experienced minds.

Thanks,
Chad Eller

I don't think you can wait a year--you may want to get www.cineform.com to help out.

heath

My understanding of the Z1 is that if you record 60i HDV and use in-camera downconversion to NTSC DV, the firewire output meets the technical specs for NTSC SD DV: 480x720, 60i, 4:1:1 color, and field order lower field first.
I have not read any posts of problems with, or need to switch field order when outputting NTSC DV.It should just capture into your system like any other DV source.
Bob

HDV-DV from Camera, Premiere custom preset
 

Through trial and error I've discovered the best preset settings (to my knowledge) to create .m2v files if you are going to downconvert HDV-DV from the camera into Premiere is:
NTSC 16:9, Quality=high, 2 pass VBR, field order=LOWER
Premiere does not have a high quality 16:9 preset configured so you have to create a custom preset to get these settings.
I still dont know about bitrates and have left the default settings:
Min bitrate= 15000 low quality
Target bitrate= 60000 med quality
Max bitate= 80000 high quality
M frames = 3
N frames = 15

I really want to try the capturing with Cineform's Aspect HD method recommended by Steven White, however I'm currently financially way over my head. Sony offered no interest, no payment till May 06 on a FX1, an offer I couldnt refuse. I've been making some money but I've been spending it on accessories (Bogen 501 head, Beachtek XLR box, Audio Technica shotgun, etc.) I've got to stop spending on goodies(i.e. Aspect HD) until I get the camera paid for (27% interest kicks in in May).
If anyone has any feedback on the bitrate and M, N frames I would appreciate it.

Wow, seems like we're jumping through a lot of hoops here trying to achieve the simple result of converting HDV material to MPEG2 output. I'd be interested to know if this is an inherent problem for all editing software due to field order issues between HDV and MPEG2, or simply a weakness in the standard encoding options available in Premiere Pro. I can try running some encoding tests from an HDV timeline in Edius to see how well that works; anyone here using Vegas, Edition, Ulead or Final Cut Pro want to try it in those programs?

My preference is always to use the simplest workflow which yields acceptable results. If I can capture and edit HDV and encode directly from the timeline to MPEG2, that's what I'll do unless some other solution is visibly better to casual observers.

Huh? HDV is MPEG2... the conversion that appears to be causing some concern here is the 'down-conversion' from HD 1080i to 480 (576 for PAL). From my personal perspective, your observation of "a weakness in the standard encoding options available in Premiere Pro" certainly proved the case with handling HDV.
I now use Vegas for all my HDV material (in conjuction with ConnectHD), as HDV support (that's not capture support!!) was added. Rendering Options are also a little easier to access and understand in Vegas... but that's just my opinion.
A CFHD/m2t/mpg HDV clip loaded into Vegas will be correctly recognised.

Of course, producing a DVD compliant MPEG2 is a different matter... The best settings for rendering DVD compliant MPEG2 have been covered many times on the forum, and a search should yield what you're after.

As long as the encoder knows that the source is upper field first it shouldn't matter whether the mpeg is made upper or lower...most modern DVD players don't care if the mpeg it is upper or lower....really old ones needed upper but my guess is 98 percent of the players now don't care.

Steve: you listed several steps you're going through to convert 1080i HDV to 480i MPEG2, which in theory should be a simple and straightforward process. That's what I'm trying to understand: is there any software which can do this effectively from the timeline without a lot of intermediate steps? Like I said, I can test this in Edius but would like to know what results people are getting in other software.

I can encode from the premeire timeline from cineform to dvd mpeg2 using Procoder or Main Concept, both of which let me choose field order from there plug-in menus.

My problem is that the 16x9 widescreen dvd output looks very soft compared the cineform orginal.

I think I may be having colorspace issues, but I do not know what the FX1-cineform-premeire colorspace workflow is, or if it is even consistent.

I do not want re-edit in vegas just to fix this - or does a program like automatic duck exist to help one load a premiere project in vegas?

I thought I'd put to rest (at least for me) the software down-conversion vs. in-camera DV conversion debate by doing a test. Here is the frame I chose:

http://s94963366.onlinehome.us/HDRFX1/softvincam.bmp (2 MB)

The top image uses the software down-conversion I outlined earlier in this thread:
- Capture 1080i HDV, in this case to Cineform Large via HDLink from AspectHD 3.1*
- Place 1080i HDV in a 59.94 fps timeline in Adobe After Effects 6.5 Pro
- Interpretation Rules: set upper fields first, deinterlacing at best quality
- Render to 720x480p60 uncompressed file.
- Place uncompressed file in 720x480p60 timeline
- Render to 720x480i60 uncompressed (click field render - lower field first, conform to 29.97 fps).

The bottom image was acquired using the in-camera DV down conversion.

My analysis

Looking at the images in the interlaced format presented doesn't show much obvious difference... most notably, you will see a slight difference in contrast and saturation - I believe this is due to the color space conversion YUV.709 and YUV.601. This image is a 24-bit RGB bitmap. I did not perform any colour correction.

The main flaw with the in-camera down conversion is the re-compression of the image to the DV codec. The DV codec both spatially compresses the image, as well as resamples the colour space from the essentially 4:4:4 available to 4:1:1. This generational loss is easily apparant in the bottom image.

For example, the neck of the goose is in essentially a black and white part of the image, and is a very sharp transition. If you zoom in on the neck in the top image, it looks quite clean. Zooming on the bottom image reveals considerable mosquito noise around the edge of the goose's neck. It is noticable in both the black of the neck, and the grating in the background.

To demonstrate how offensive the colour resampling is, have a look at some of the longer blades of grass in front of the goose. On the top image the green is clearly confined to the grass. On the bottom image - in particular, in front of the goose's body, you can see the green spilling over onto the goose. Again, this is everywhere, but you have to look for it.

Now, imagine that you're taking this image to an MPEG-2 TS for DVD. At this point, you will re-sample the colour space to 4:2:0 and do more spatial compression. It seems obvious to me that the upper image has a lot more information to provide to the compressor, and ought to result in a significant improvement in image quality - whether perceivable during real time display or otherwise.

-Steve

*Yes, this is an inherently lossy compression. But compared to what the DV codec did, it's irrelevant, and it is significantly easier to work with. If the reader wishes, they can prove to themselves that the Cineform solution is worthwhile.

**Note that no intermediate renders were to compressed formats. All images were rendered to 24-bit bitmap sequences.

Steven is dead on with his remarks. That is a lot of work to get to a conversion from 4:4:4 to 4:2:0 but technically is the very best method. I do not go the uncompressed route... Edius using HQ to make a dvd basicaly goes from 4:2:2 to the DVD 4:2:0 and there is no conversion down to DV spec(4:1:1 NTSC and 4:2:0 PAL) that occurs. Because the resolution of HDV is twice the resolution of DV there is more than enough information to create 4:4:4 though.




Horizontal Vertical NTSC
4:4:4--- Y 720 480
Cb 720 480
Cr 720 480
4:2:2--- Y 720 480
Cb 360 480
Cr 360 480
4:1:1--- Y 720 480
Cb 180 480
Cr 180 480
4:2:0--- Y 720 480
Cb 360 240
Cr 360 240
In case of HDV (1440x1080):

4:2:0--- Y 1440 1080
Cb 720 540
Cr 720 540
If we just divide by 2 the horizontal and vertical resolution of HDV signal then we get the result below:

Y 720 540
Cb 720 540
Cr 720 540



which is where the 4:4:4 can come from.

I have a tape question, does MPEG IMX or Digibeta support 4:4:4?