HDV --> SD DVD Workflow?

[hhansard]

Dont want to sound stupid or naive, but probably going to come across that way

I have shot [underwater] dv for years, and moved to hdv in 06. Since I have an end-to-end blu-ray setup with a broadcast scalier, my hdv footage looks great at home.

I was asked to take some of my hdv footage and burn an sd dvd for distribution. I took an edit of 1440x1080 and exported in PP-CS3 [movie] [Sorenson] as 720x480dv, no recompression highest quality (26G for a 30minute vid), etc, etc.. and imported it into a new PP3 d1 project. Burned the DVD again highest quality

I looked at the results on an sd 4:3 monitor terrible much worse then my old native dv footage. I expected to loose quality, but assumed that the scaling algorithm would smooth, and I would end-up with something acceptable --- but it does not look like it it looks like pixels are just dropped with no interpellation at all. I mean if you never saw the original footage, you might let it pass, but having seen the original footage, you can tell that the compression has killed it. I know that this is like a 2 (maybe 3) generation dupe, but I have access to the original pixels and would have assumed with minimal recompression/expansion the results would be as good as 1gen dv but I cant seem to get there

Can anyone point me to a good workflow [or some settings] to take hdv footage and cut a decent quality sd dvd using the production suite??

Thanks in advance,

Hugh

[Curt Wrigley]

Step 1-3 are simply to capture your clips as hdv.

Step 4 means; close the hdv project, create a new ppro project using the ntsc dv 16:9 preset. Then import the hdv clips into that project.

If you email me at:

info (atsign) wrigleyvideo(dot) com ill give you my IM so this can go faster. I have great interest in your results.

Curt Wrigley

[hhansard]

Got it, on to 5

[hhansard]

Are you sure about default settings in #8 -- my deafult is recompress... I'll switch to e-mail...

[_hconant_]

I know it takes a little longer, but I'm sure there are a number of us that have a significant interest in this thread and the outcome, so if you guys could at least keep the meat of it in the post it would be greatly appreciated. Thanks.

[hhansard]

Sorry, there have been questions going back and forth about settings... Renders in progress...

Despite dual quad core and fiber channel, this stuff sure is slow my second fiber channel for scratch is arriving on Tuesday.

About another 9 minutes then burn, then broadcast HW...

Will publish ...

h

[_demo_crat_]

for orders sake -
try rendering out to
a)blue ray
b)wm9 (at best settings)
and compare

[_hconant_]

Hugh. Was wondering how you made out.

[Dan Isaacs]

Hi all ---

I am surprised that someone had trouble with the DebugMode FrameServer plugin! I've been using it for years (even in Premiere 6.5) without issue. I really suggest that you try to get that working. It circumvents the need to render a big intermediary .AVI of your sequence when all you want to do is convert to another format.

AviSynth is god! Install it and grab some of the free plugins (notably SmoothDeinterlace -- it's the best deinterlacer I've ever seen).

Assuming you've output your dummy frameserver .avi (or even a big uncompressed .avi, for that matter), open up Notepad and create a file called "hdv2dv.avs".

Enter the following in your blank notepad document:
---------------------------------------------------

avisource("your_filename_here.avi")
converttoyuy2(matrix="pc.709")

# uncomment the next line if you want 4x3 instead of 16x9
# Crop (168, 0, -168, -0)

# uncomment the next line if your input is full-range RGB
# levels(0,1,255,16,235,coring = false)

SmoothDeinterlace(tff=true,doublerate=true,lacethresh=24,staticthresh=40,staticavg=85,edgethresh=50,blend=true,showlace=false)
bicubicresize(720,480)

# for lower field first...
SeparateFields().SelectEvery(4,1,2).Weave()

# ... or uncomment the line below for upper field first
SeparateFields().SelectEvery(4,0,3).Weave()

Limiter(16, 235, 16, 240)

# uncomment the next line if you need RGB output
# converttorgb32(matrix="rec709")

---------------------------

And there you are... Open the .avs file in VirtualDub and render it to a lossless (or not very lossy) .avi or -- better still --- use an AviSynth compatible MPEG2 encoder (CCE, for example) and drag the .avs file into it.

This example assumes interlaced NTSC input/output, but modifying it for a PAL and/or progressive frame would be pretty easy. Tweak the deinterlacing settings and levels to your needs. Also try some of the avialable noise reduction plugins (prior to deinterlacing) -- this can really help.

Try it out and let me know how you fare.

-- Dan

[Dan Isaacs]

Oops. I just noticed a small typo in my script. Also be sure to download and install the SmoothDeinterlace plugin.

Here's the corrected version:

--------------------------------------------------------------------

avisource("your_filename_here.avi")
converttoyuy2(matrix="pc.709")

# uncomment the next line if you want 4x3 instead of 16x9
# Crop (168, 0, -168, -0)

# uncomment the next line if your input is full-range RGB
# levels(0,1,255,16,235,coring = false)

SmoothDeinterlace(tff=true,doublerate=true,lacethresh=24,staticthresh=40,staticavg=85,edgethresh=50,blend=true,showlace=false)

bicubicresize(720,480)
SeparateFields()

# for lower field first... or use SelectEvery(4,0,3) for upper
SelectEvery(4,1,2)

Weave()

Limiter(16, 235, 16, 240)

# uncomment the next line if you need RGB output
# converttorgb32(matrix="rec709")

[hhansard]

What results did we get? That is a good question

Some experiments yield definitive results; some are directionally interesting and yield the need for more experimentation

Unfortunately I believe that we landed on the later... I dont actually have access to the equipment that I need to be definitive until later in the week

Curts support is to be thanked We traded e-mails as workflow steps progressed and questions arose

First of all, the result is compromised: I ran the SD DVD through a broadcast scalar to an HD 1080p broadcast monitor, so there [is probably] a difference in color space I never remember how all these standards map I am not a pro, nor do I play one on TV.

Curts workflow produced better results then I received in previous experiments

The results were interesting... I have a hypothesis as to what is going-on, and Curt has suggested an experiment to test it

So as they say, back to the dungeon I should have the follow-up experiment completed by tonight

Hugh

[_hconant_]

Hugh, thanks for the updates. Dan, thanks for your continued input. I am still working on your solution (and maybe that's the one I'm looking for). On a slightly different note, and let me know if it should be a different thread; I had posed the question of, if one was willing to spend the money, is there a combination of camera (and its corresponding high def format), LNE hardware, and NLE software, that makes the process of going from HD to SD, or HD to any other format for that matter, a relatively effective process with a predictable, reliable, and high quality outcome. And by HD I am meaning simply high definition, and do not mean to imply a particular format. And I don't intend to be repetitive, however, it has been implied, often, that the source of the problem is the inherent problems associated with the compressed mpeg2 format coming from the lower end HD handycams.

[Dan Isaacs]

Hi hconant ---

I hope you are able to achieve the results you're looking for with my method. Let me digress a bit before I answer your second question...

I've been working with pro-sumer NLEs since the early 90s and I've used many different capture cards and formats (FAST GmBH, Broadway, Canopus, Matrox, etc.)

Currently, I'm using the Matrox RT.X2. I've got a pretty long list of complaints about it, but it's all right for my purposes. All of the footage I use now is shot in HDV but I do the editing in SD 4x3 (DV). I convert all the HDV footage to DV prior to editing.

I like the interface of Premiere, but I've always had issues with its technical "precision" (or lack thereof). That's how I came to learn AviSynth. Expense-wise, you can't beat it and as long as you are working ona Windows system you can use it to aid in just about every kind of conversion. All you need is a good MPEG2 encoder that can read .avs files. I recommend CCE SP2 (Cinema Craft Encoder) as it can handle YUY2 data directly.

There are many other systems and software around but if you're comfortable working in Premiere then stick with it. Use tools like AviSynth, VirtualDub and CCE to compensate for Premiere's weak points.

HDCAM is better than HDV and Avid also has their uncompressed HD format but I don't know too much about either. If you're looking for bang-for-the-buck, however, I'd spend my money on a good "professional" HDV camcorder and some nice lenses.

Once you've got it established, the AviSynth method will yield very nice and consistent results for HDV -> SD conversions.

[_tim_kohn_]

100% agree. I have same camera, and used the exact same workflow last year when capturing many hours of ballet recital video that was shot in HD, and multi-cam edited in HD. Then used the procedures described to achieve very good quality SD DVD's. We produced nearly 800 DVD's from six recitals, and have gotten great feedback from our customers commending the quality. It took me awhile back then to figure out the workflow, but it did produce results.

[Dan Isaacs]

The best thing about the DebugMode -> AviSynth -> CCE workflow is its flexibility. You can specify exactly what kind of processing takes place prior to conversion (and in what order it occurs).

AviSynth will allow you to convert colorspaces and luma ranges, framerates, PAL <-> NTSC, interlaced -> progressive with nearly infinite options for how each step is accomplished. A takes a little while to learn, but you can find help and examples on the internet very easily.

Assuming you're running on a decent computer system, the speed of processing is also fantastic. AviSynth and many of the available plugins are well optimized and run with little overhead.

I'd also like to recommend QueEnc: A freeware MPEG2 encoder. While not as sophisticated as CCE, it is totally free and also provides excellent results.

IMHO, Adobe has a lot to learn from the freeware developers :)

[hhansard]

Phase II experiment complete

Sections:

The workflows in the experiment
Experiment design
The Hypothesis that pushed the workflows and experiments
The Results

This is a very long post and represents many hours of work and the support of many

I wish thanks in advance.

I may have to break this down into multiple posts to manage the size

--------------------------------------------------------
The workflows.
--------------------------------------------------------

I abandoned my workflows in favor of Curts (see previous post). He is getting the directionally best results using an all-Adobe workflow and represents the quickest way to get my project in-the-can. I need get this in the can, so any workflow that gets me closer is a direction that I am willing to investigate.

Workflow #1:

1. Shoot in 60i
2. Create new PPRo project with HDV 1080 60i preset; dont change anything
3. Import HDV clip
4. Create PPRO SD 16:9 project (dont change anything.
5. Make sure the general preferences are set to NOT scale to frame size
6. Import the HDV clip and drop it on the timeline.
7. Change the scale setting in Motion to 45%
8. File/Export Movie (default settings of MS DV AVI)
9. Launch Encore/Import the exported clip as a timeline. Burn with default settings.

You could export to Encore from PPRO in step 8; or frame serve, but I use the workflow i detailed out of habit. There is negligible loss exporting to DV AVI one time.

After Workflow #1 was run, results were obtained that required further investigation.

This produced a suggestion from Curt for Workflow #2:

1. Shoot in 60i
2. Set vtr to widescreen SD, let VTR do all conversions (aspect ratio, scale, any color mapping (if needed), etc, etc. In short, replace steps 2 7 with a direct hardware conversion and bypass PPs render engine. It will also be working in avi and not mpeg)
3. Import directly into SD widescreen project
4. Launch Encore/Import the exported clip as a timeline. Burn with default settings.

After this was run, a further workflow was needed due to some color shifting noted.

1. Shoot in SD widescreen
2. Directly import
3. Launch Encore/Import the exported clip as a timeline. Burn with default settings.

Now this part is a bit flawed, because it used a different CAM, but the results of #1 required viewing of some legacy footage in the same project to answer a few questions about color shifts

--------------------------------------------------------------
The experiment
--------------------------------------------------------------

The reason and structure of the experiment will become clearer when the hypothesis is presented below. The hypothesis was developed after the results of Workflow #1 being run.:

Take 3 separate 1 minute clips of varying subject matter.

Clip #1: This clip is fairly even in tone and luminosity. There are color/tone ranges within the clip, but they are not contrasty. The scene is essentially static with only a little movement by the subject.

Clip #2: negative space separation through color contrast but too much. There is more movement of colors against contrast backgroundbut not too much contrast.

Clip #3: Negative space management through both color and luminosity contrast. The subject is tracked across a deep and highly saturated blue background. The subject has bright yellow, bright orange and jet black. The colors are interacting with the background.

All clips are NTSC broadcast safe as verified on a hardware scope.

Take all three clips through all three workflows: 9 clips, each of 1 minute

Title each clip with a number in the lower right corner to identify clip not the workflow, just the clip. Randomize the experiment to minimize bias.

Create 3 sequences in 1 project. Put the three clips processed with workflow #1 in the 1st sequence, and then do the same with the rest.

Have wife come in and scramble clips then output to Record which clip is which Encore and burn

Obviously, Clips in workflow #3 are identifiable; they are not exactly the same as the others, but very close in composition and exposure. Same subjects, just 400 miles apart So well call this the control group Yes, its a bit flawed, but I needed to answer a question about color accuracy, and this seemed to be the best way to do it

Develop questionnaire to grade the clips 1 to 5

I do not have access to calibrated SD equipment for a couple of days, so the definitive results will have to wait until then. Had to compromise a bit on running the output, but should be directional.

Take DVD into town and bother local photo studio again. Explain that these are images processed in a verity of ways and ask them to watch the DVD once for reference, then once to grade the results. 2 customers came in, so we now had an n of 5

Collect results from high-end consumer equipment.

Show results output from original hdv vtr on their other consumer 1080 monitor.. Collect qualitative results

Talk a friend into watching this on through a broadcast scalar to a 1080 broadcast monitor.. This is also slightly flawed but still, could get direction.

------------------------------------------------------------------
The hypothesis
------------------------------------------------------------------

The experiment reveals the hypothesis. In evaluating the results of the original experiment involving workflow #1, I began to notice a pattern The more dynamic the footage the more the interlace artifacts. With my previous workflows, the artifacts were more generalized, but with Curts workflow they became more localized to the boundaries where high color contrast occurred. The more the movement between the colors, the more the artifacts .

Several posts suggested the problem is with my footage The implication being that the exposure, saturation, etc are not of sufficient quality to get good output. But that was not the case A broadcast monitor and scope reveal all

But turn this idea around suppose the issue is not my footage, but my footage is causing issues with a PPro render it may be my footage, but not in a way that folks have suggested Suppose that you have footage that is "good" and broadcast safe, but contains color combinations that PP has a difficult time managing?? What would happen during scaling, encoding color space mapping, etc, etc. What result would you get??? Would they be acceptable???

It might also explain why some are satisfied with their results and some arent might depend on what you feed the system

So, workflow #2. And then ran an experiment to compare side-by-side with #1.

Result: Workflow #2: eliminate the PP render, scale, aspect conversion, color mapping, etc, and go right to Encore and see what happens It should be mentioned that a variant of this was proposed by a production pro in this thread two weeks ago..

----------------------------------------------------------------
Results:
----------------------------------------------------------------

In the simple clip, folks were pretty much 50/50, but after we moved-on, the results started to shift toward workflow #2. In short, the more complex the footage, the more the workflow #2 started to out-perform workflow #1.

Color: There is also a noticeable color shift towards greens in workflow #1 that is absent in workflow #2. That was the reason for workflow #3 to see which color representation was more consistent. This became even clearer when the raw HDV was played

Halo artifacts: The moving subject (in clip #2 a green turtle is moving against a blue/green background think I said something about underwater some posts back) has a clear halo. You dont pick-up in the first viewing, just a basic sense of softening, but when you watch the vid several times closely, you see what is going on.. This is str

[hhansard]

OK, I guess the post was too long -- sorry about that...

I will start again at the Results boundry:

----------------------------------------------------------------------
Results:
---------------------------------------------------------------------

In the simple clip, folks were pretty much 50/50, but after we moved-on, the results started to shift toward workflow #2. In short, the more complex the footage, the more the workflow #2 started to out-perform workflow #1.

Color: There is also a noticeable color shift towards greens in workflow #1 that is absent in workflow #2. That was the reason for workflow #3 to see which color representation was more consistent. This became even clearer when the raw HDV was played

Halo artifacts: The moving subject (in clip #2 a green turtle is moving against a blue/green background think I said something about underwater some posts back) has a clear halo. You dont pick-up in the first viewing, just a basic sense of softening, but when you watch the vid several times closely, you see what is going on.. This is stronger in workflow #1 then #2, but present in both.

Interlace artifacts: there are noticeable interlace artifacts at the boundaries of strong color contrasts These are more present in workspace #1 then #2, but still present in both

The raw vid There is not a question, workflow #1 and #2 did not even come close to the raw vid on broadcast equipment.. I spent hours comparing the results..

So; workflow # 2 works better then #1??? Well I think it depends on what you ask the software to do??? Is this significant?? Should be directional

With these workflows, I could not get there. Not with my footage

I know it can be done. I have seen extraordinary SD DVDs shot with HDV, then pushed down But not with these workflows. But can it be done with these tools at all????

Now, that is a question

Hugh

[Curt Wrigley]

The reason i suggested you try workflow 2 with the same footage is it sounds like your footage is NTSC chalanged. It can be very leagal on scopes and withing exposure and color boundrys and still be a ntsc problem. Try red titles in yellow for example.

So; if workflow 2 is still a problem, I submit that even if you shot it in SD you would be seeing the halos. Some of you problems may be downconversion issues, but some of them may just be ntsc challanges. Dont discount it.

Curt

[Dan Isaacs]

Hugh --

I really apprecaite the effort you are putting into evaluating these different workflows. Your experience will serve as a benefit to everyone in this forum. However...

Soft scaling algorithms + no noise reduction / filtering + cheapo deinterlacing + inaccurate color mapping + unknown handling of illegal colors + mediocre MPEG2 compression...

...equals what? Yep. You got it. The "all-Adobe" workflow :)

I would be honored if you gave my method a shot and put it to the test. I think you'll find that it not only provides better quality (and control) but will allow you to get stuff "in the can" faster.

Premiere is a great editing program. Conversion is not (and never has been) its strong suit -- in regard to both quality and performance.

Thanks again, and keep up your experiments!

Regards,
-- Dan

[Curt Wrigley]

Dan, what you are disregarding is, he has the same problem with SD footage. I do not find adobe's mpg encoding to be mediocre. The other issues you metioned are not even in play in the second workflow (SD footage)

[Dan Isaacs]

Hi --

Thanks for the reply. I'm glad this is turning out to be a lively discussion :)

Well, I may or may not be disregarding it. True, I guess it is unfair to blame everything on Adobe products (in the case of workflow #2). I was being overly emphatic -- and joking a bit, too.

But here are a few considerations that do apply to Workflow #2 (sorry in advance for the lengthy post):

1.) SD (DV) footage is CCIR601, while (by default) DVD is CCIR709. Is the Camera/VTR outputting 601 or 709? Is the Adobe MPEG encoder flagging this correctly -- or --- are the DV .avi's being converted to RGB and then back to YUV (YV12) for MPEG2 compression by the Adobe encoder? It would be hard to determine without input from Adobe themselves.

2.) I imagine that not all Cameras/VTRs are created equal in regard to their ability to transcode. HDV footage, while higher in resolution, is more prone to artifacting than DV. What kind of noise reduction, deblocking, de-ringing, etc. (if any) is being applied by the VTR prior to conversion? Maybe none is needed (depending on your source) but if it is, you're stuck. This is true for both workflows #s 1 + 2.

3.) DV and HDV both allow luminance ranges of 16-255 (super whites). Broadcast (and DVD) standard only allows 16-235. Are illegal values clipped? Are they scaled? Is the whole range scaled up to 0-255 and then clipped to broadcast range? Again, keep in mind 601 <-> 709 and RGB <-> YUV conversions. How are "hot" colors handled? There are many possible combinations of color/luma transformations taking place -- none of which are transparent to the user. Also true for both workflows.

4.) If shot in 60i, the VTR needs to perform deinterlacing prior to scaling the video. There are many methods for doing this and I doubt that many mid-level HDV decks and camcorders perform this step with great fidelity. Even the best deinterlacers can fail on certain material. This is why it is important to be able to tweak these values to suit your needs. Furthermore, if the VTR is to output standard NTSC DV (interlaced, lower field first) then you must do some rather fancy processing (doubling the frame rate and re-weaving the fields is a rather good way, assuming your deinterlacing is performed well.) Let's say you want to output progressive frame SD from an interlaced source: Is the "extra" field chucked away or is the pair blended together? What method(s) does your VTR use? Maybe this is documented for some cameras, but certainly not all.

5.) What about chroma upsampling? I won't go into depth about it, but this is another unknown factor. HDV is 4:1:0 (as is DVD) but NTSC DV is 4:1:1. Is the chroma being nicely upsampled by your VTR? Does this occur before or after scaling? It does make a difference. Also: How does Adobe MPEG encoder handle the 4:1:1 -> 4:2:0 conversion?

6.) Let's say you have a finished HD timeline in Premiere. How would you apply Workflow #2? You'd have to render your timeline back to the VTR (and I've heard some horror stories about that process alone). At best, you'd still have recompress the content to HDV MPEG2. Now revisit all the aforementioned pitfalls, caveats, and unknown factors of exactly how the Adobe MPEG2 encoding works and we've multiplied our problems. HDV MPEG2 encoding can also be very slow. You have to add the time of rendering, exporting to tape, and recapturing. Even in a best-case scenario, you've already compromised the integrity of your video by doing an interim lossless -> HDV -> DV conversion. All unnecessary, if you ask me: Just frameserve the full-quality frames from Premiere to a capable video processor (AviSynth) and MPEG2 converter (CCE, QuEnc, etc.) and you're done.

7.) I refer to Adobe MPEG encoder as mediocre because it does not provide many options and has many "unknowns" as to how it operates. I'm sure it does a nice enough job on the compression itself if you feed it exactly what it wants (and how do we know this?). Consistent results seem to be the issue here. It is also not a standalone application and doubtless uses more system resources than it needs to encode the MPEG2 stream.

[hhansard]

Wow, who said I had a problem with native SD foortagle in a native SD workflow --- I think my posts have been fairly clear ---- that is what started this in the first place...

Yah, I have problems with #2, but that is not native SD... It is running vid through a hardware process.. I used a 35U to map color, scale, change aspect ratios, and god knows what else... so your milage my vary based on the footage and hardware used...

Do I get results from #2 that are as good as native SD??

Don't know --don't think so... but can't verify until I get access to the right equipment..

I have access to a broadcast facility on Wednesday. I'll stop by with raw vid and verify NTSC legal with a friend who is a a broacast engineer. I will then seriously impose on him to see what he can do with the footage... It might be fun to see what happens if you throw a zero or two at the problem...

Dan, I will try your workflow... afterall; in for an ounce...

Hugh

I'll try Dan's workflow.

Hugh

[Dan Isaacs]

That's great, Hugh...

If you need any help finding / installing / configuring any of the freeware tools I've mentioned (DebugMode frameserver, AviSynth, SmoothDeinterlace, VirtualDub, QuEnc) I'd be glad to help.

Absolutely right: switching your HDV camera to DV output is not native SD at all.

Not for nothing, but I've done a lot of video conversions in my life. I am not a broadcast engineer and I've never had much access to super-high-end equipment. To the contrary, I've had to learn (the hard way) how get acceptable results on a low budget.

I also got tired of trying to guess what a partiular program or device is doing (or not doing) with my video. This is how I came to learn AviSynth in the first place. Absolute control, leaving nothing to chance, then just feed it to the encoder for a "straight" encode.

AviSynth is also great for doing NTSC <-> PAL conversions. Assuming you have a HDV 60i "master" sequence in Premiere, you can quickly and easily generate both PAL + NTSC (interlaced or progressive) SD DVD content from it.

Once you've got my suggested (simplified) workflow implemented (and you've learned some AviSynth basics along the way) I can show you other ways to monitor, optimize and control such conversions.

[Curt Wrigley]

Hugh,

Just to clarify. I suggested workflow 2 to eliminate all the scaling being done in PPPRO; having your camera do the scaling. Thats the closest we can come to comparing SD NTSC footage of the same scene. Most cameras do a pretty good job of this conversion. Since you have the same or similar problem burning this SD footage produced by your cam, then we can remove a ton of variables from the table.

Much of your problem has been; you are testing and adjusting dozens of variables at once and declaring PPRO cant scale and burn HDV footage. Much of the problem was simply messing with too may variables. When your problem may in fact be footage that is so ntsc challenged that no workflow will provide the results you want.

CS2 did indeed have significant scaling problems. CS3 is very much improved such that many users use ppro or encore to scale hdv footage routinely with excellent results.

[hhansard]

Dan,

I am starting to implement your workflow... This is going to be a learning curve -- I'll call it: experiment #3...

I have started to review the materials, and will try some small sub-experiments to learn the tools... I don't want to clog this thread with the obvious settings questions that will arise... If you don't mind, can you e-mail with an e-mail address for a couple of the questions...

hugh@hansard.org

Thaks, hugh

[_Shawn_M_Lam_]

No wonder my HDV to SD DVDs look like crap. I think I'm going to just shoot 16x9 SD for now until there is an easier solution.