DVD Studio - HD to SD hell!

[Steve Rotter]

hey guys, what's up? i am sure compressor works really nice but i almost drove myself mad, working on all these different configs for 3 weeks straight. what i did, and it works great...for me anyway....and looks nice too!.....

make sure the timeline settings for final cut pro 7 are set to apple pro res 1920x1080 30p.

i shot footage in 1920 30p on the canon hv 30 and xh a1.

edit in final cut pro

export...file / export quick time. i export as quicktime movie and select the custom option for codec as apple pro res 1920x1080 30p, which is the same the timeline is set for. this looks great.

i bring this hi rez .mov file right into idvd (stopped using dvd studio pro.) idvd is even faster, looks great, and does the encoding right. i set the options for idvd for professional quality. idvd does the encoding on its own. took about 4 hours for 45 minutes footage.

idvd did a nice job. my titles have no pixelation and things look great! the only thing i don't like is with faster motion...people running past the camera on tripod...there is some jerkiness. i think this is attributed to filming in 30p. if i filmed in 60i, that wouldn't be there but i CAN'T STAND THE CLEAN LOOK....looks too much like a video camera at a kids bday party. what if i filmed in 30p but captured in 60i and exported in 60i? i don't want to cause a problem.

right now i'm assuming you have to capture and export based on how you filmed....30p or 60i, from filming, all the way to the end.

currently i'm working on blu ray on the mac. i will experiment with exporting a qt .mov as i am now, and importing into adobe encore cs4 for the enode. the H.264 looks outstanding so i will select that in encore.

[Randall Leong]

Originally Posted by Alex Kuzelicki

Hey there,

I'm in the midst of a bit of a nightmare (of my own making) at the moment, and hoping some tech-savvy HD-head out there can offer up some advice.

We have some great DVCProHD 1080i (PAL) footage - a kid's show - that we want to burn to standard definition DVD. Unfortunately, the quality of our initial attempts in DVD Studio Pro came out pretty crappy. Personally, I couldn't believe how bad the image looked compared with our beautiful HD source footage.

Despite the knowledge of my DVD Studio Pro guy, and due to some time restraints, we took the HD footage to a 'pro' post-production house - thinking they must know something we don't. But their results weren't much better... and still cost us a big chunk of money! (Later, we found out they just used DVD Studio Pro as we did)

So, now, we're still where we started. We have great-looking HD footage but can't get it looking what we consider 'acceptable' on a standard definition DVD. Surely, this is possible? I've been fed a bunch of opinions (eg. from the post-house especially) that compressing HD for SD-DVD delivery 'just turns out that way'. I can accept a certain amount of degradation but, by comparison, the end result is truly atrocious!

I mean there must be a way to get nice clean pictures from HD to SD-DVD... I see them every day at the video store!

Can someone enlighten me?

Thanks in advance,

ALEX

I know that the OP has not posted since this past May, but I am quoting this to show you that kind of result is all too typical if you rely solely on the NLE for the downconversion of interlaced video content. The NLE's built-in encoders do a terrible job of downsizing interlaced videos: They either deinterlace everything (which throws away one-half to three-fourths of the total video information even before downsizing) or nothing (which nearly always leaves obvious, and often severe, motion artifacts). And their resizing does not use the most appropriate alogarithm for the conversion of square pixels to non-square pixels.

On the other hand, the nice-looking results done on professional movie-studio SD downconversions from HD are done with astronomically expensive hardware-based motion-adaptive deinterlacers--priced in the order of tens to hundreds of thousands of dollars per system. And the video store that you visited for those "nice" HD-to-SD conversions might simply have been playing SD media (standard DVDs which, once again, had already been authored by a professional motion picture studio) or playng HD media on an HD player that's connected to an SD monitor (and the HD player will likely do a better job of this downconversion than an NLE by itself).

And there are freeware software programs (such as AVISynth/VirtualDub with the appropriate freeware plug-ins) which can produce HD to SD downconversions which look nearly as good--but they are very slow to process the videos (they take as long as 10 to 12 times the original length of the video clips themselves to do their work).

By the way, before I discovered AVISynth and VirtualDub and their plugins, I used to use an NLE alone in downconverting 1080i HD to 480-anything SD. But my results ranged from the sharp but artifact-laden videos coming out of Vegas and Ulead/Corel to the soft and often blurry videos coming out of Pinnacle Studio. Neither result proved to be what I consider 'acceptable', so I kept searching for an effective but not-too-costly solution.

[Pavel Tomanec]

Originally Posted by Steve Rotter

idvd did a nice job. my titles have no pixelation and things look great! the only thing i don't like is with faster motion...people running past the camera on tripod...there is some jerkiness. i think this is attributed to filming in 30p. if i filmed in 60i, that wouldn't be there but i CAN'T STAND THE CLEAN LOOK....looks too much like a video camera at a kids bday party. what if i filmed in 30p but captured in 60i and exported in 60i? i don't want to cause a problem.

right now i'm assuming you have to capture and export based on how you filmed....30p or 60i, from filming, all the way to the end.

Hi Steve,

the look depends also on the frame rate, shutter angle and aperture setting. Also you may need to do additional research into plug ins that help with curves and saturation closer to film feel. I strongly suggest not to shoot in 30f for narrative projects, but in 24f if you wish to have a bit of "film look"

Look into Nattress plug ins I believe there is a conversion from 30f to 24f. People use them and are happy.

Best.

[Steve Rotter]

thanks Pavel, i will do that. will panning be even more jittery if shooting in 24p? i like the smoothness of 60i but it's too clean...like video cameras from the 90s.

[Sorin Pricop]

Hi guys,

Just a thing ...

I have had the same problem - but my solution is ... how can i put it ... complicated!

I've imported the footage from the P2 - done the project - exported to tape - from tape imported it as DV(Wide) (downscaling from the camera [FX1]) ....
The rest ... Encore ...

Result = Pretty nice looking video ...

Regards,
Soreen

[Greg Boston]

It was mentioned a couple of times in this thread, but as a rule, the encoder that DVDSP uses will not produce the highest quality DVDs. There are third party encoders that do a much better job.

The big, hardware based Cinemacraft encoder is now available in software form and is considered to be the cream of the crop. Other encoders that cost money come from Sheer Video, BitVice, etc. These encoders will all provide results that surpass what DVDSP can produce.

Sorin also touched on something that used to be discussed here a lot when converting HD to SD. The camera can often do a better job than software!

-gb-

[Robert M Wright]

Originally Posted by Randall Leong

And there are freeware software programs (such as AVISynth/VirtualDub with the appropriate freeware plug-ins) which can produce HD to SD downconversions which look nearly as good--but they are very slow to process the videos (they take as long as 10 to 12 times the original length of the video clips themselves to do their work).

I'm not sure why your HD to SD down-conversions take so long with AVISynth or VirtualDub, unless perhaps you are using a really old computer.

Using VirtualDub, can produce excellent down-conversions from HD to SD, and (with a reasonably modern PC) isn't that slow. I wouldn't do anything fancy (or time consuming) for deinterlacing 1080i60 source to 30p - simply drop a field. Dropping a field gets you to 540 lines (progressive), before resizing. Then simply use Lanczos resizing to get to 720x480.

Another way to do deinterlacing, is simply resize whole frames, telling the resizing filter that the source is progressive (don't check the box for resizing interlaced source). What you effectively get is a blending of fields that gives you a result somewhat akin to having shot at 1/30 shutter (if your source was 1080i60 at 1/60 shutter). For footage with very little motion, this can work very well. With low motion footage, it can reduce judder subtly and also effectively act as a temporal denoiser.

Another thing you can do with VirtualDub, is apply some visually lossless denoising with MSU's (free) denoising filter. With typical footage, on the lowest preset, MSU's denoiser will produce a result that is visually the same as the source, yet significantly cleaner for final compression. Using the MSU denoising filter will slow things down a bit, but it's not that awful with a reasonably powerful CPU. (I don't suggest using the option for GPU acceleration with that filter. It can increase performance noticeably, but can also often result in undesirable artifacting.)

Personally, I don't see much point in using a really expensive encoder, like Cinemacraft's, for MPEG-2 encoding to make a DVD. TMPGEnc 2.5 is a very low cost encoder, that produces excellent results. The Cinemacraft encoder will give you slightly better results, but not significant enough to notice at all during playback. You have to step through the footage frame-by-frame, and blow it up, to see any difference, really (awfully minor).

[Randall Leong]

Originally Posted by Robert M Wright

I'm not sure why your HD to SD down-conversions take so long with AVISynth or VirtualDub, unless perhaps you are using a really old computer.

Using VirtualDub, can produce excellent down-conversions from HD to SD, and (with a reasonably modern PC) isn't that slow. I wouldn't do anything fancy (or time consuming) for deinterlacing 1080i60 source to 30p - simply drop a field. Dropping a field gets you to 540 lines (progressive), before resizing. Then simply use Lanczos resizing to get to 720x480.

Another way to do deinterlacing, is simply resize whole frames, telling the resizing filter that the source is progressive (don't check the box for resizing interlaced source). What you effectively get is a blending of fields that gives you a result somewhat akin to having shot at 1/30 shutter (if your source was 1080i60 at 1/60 shutter). For footage with very little motion, this can work very well. With low motion footage, it can reduce judder subtly and also effectively act as a temporal denoiser.

Another thing you can do with VirtualDub, is apply some visually lossless denoising with MSU's (free) denoising filter. With typical footage, on the lowest preset, MSU's denoiser will produce a result that is visually the same as the source, yet significantly cleaner for final compression. Using the MSU denoising filter will slow things down a bit, but it's not that awful with a reasonably powerful CPU. (I don't suggest using the option for GPU acceleration with that filter. It can increase performance noticeably, but can also often result in undesirable artifacting.)

Personally, I don't see much point in using a really expensive encoder, like Cinemacraft's, for MPEG-2 encoding to make a DVD. TMPGEnc 2.5 is a very low cost encoder, that produces excellent results. The Cinemacraft encoder will give you slightly better results, but not significant enough to notice at all during playback. You have to step through the footage frame-by-frame, and blow it up, to see any difference, really (awfully minor).

I think I might have had some settings incorrect. I retried downsizing again--only this time, I separated the fields ("bobbed") in AVISynth, then resized and reinterlaced ("woven") in VirtualDub--and found that the process (on my Core2 Quad Q9450 PC running Windows) took about two-and-a-half times as long as the length of the video (five minutes of processing for every two minutes of video recorded). And that is from 1920x1080/60i to 720x480/60i, TFF (or UFF). Downsizing from 1280x720/60p to 720x480/60i proved to be significantly faster than from 1920x1080/60i to 720x480/60i because downsizing from 720p60 material does not require bobbing or weaving, only resizing and scan-conversion (dropping alternate scan lines from each frame after resizing). Plus, if I were preparing the new AVIs for standard-definition DVD, I would now make sure that I set the "output" file at YV12 (4:2:0) so that the editing and/or DVD authoring software don't have to perform their own color space conversion.

And yes, in my earlier days (when I first used AVISynth/VirtualDub), I used to drop one field then deinterlace the rest--but the resulting videos I got were all a flickering, choppy, stuttering, blurry mess (something that I never noticed on the HD 60i originals but made painfully clear on downsized SD copies). I want smooth motion, not this stuttering flickering cr-p, when my original source is native 60i video. Dropping a field only works correctly with 30p video that had been encoded in a 60i transport stream, IMHO (and even then, I would have ended up losing half the vertical resolution in the final image, effectively making it a 720x240/30p video).

And you were still thinking that 60i equals 30p. That is incorrect. The "shutter" in a 60i camera refreshes every 1/60 second--and each 1/60 second contains a different image. A camera which records in 30p, on the other hand, either refreshes only every 1/30 second ("native" 30p) or electronically splits the frame into two separate half-rez fields to match a 60 fps frame rate (30p embedded in a 60i transport stream).

[Robert M Wright]

Originally Posted by Randall Leong

I think I might have had some settings incorrect. I retried downsizing again (only this time, I separated the fields in AVISynth, then resizing and reinterlacing in VirtualDub, and found that the process (on my Core2 Quad Q9450 PC running Windows) took a little more than twice as long as the length of the video (a bit over two minutes of processing for each minute of video recorded). Plus, if I were preparing the new AVIs for standard-definition DVD, I would now make sure that I set the "output" file at YV12 (4:2:0) so that the DVD authoring software doesn't have to do as much work.

I have no idea why you are separating the fields, resizing, and then recombining them. If you want to resize interlaced footage, and get interlaced footage as a result, simply check the box next to "Interlaced" in the dialog for resizing and, unless I'm utterly mistaken, the resizing algorithm you choose will be applied to the fields individually (no need for all the extra fooling around).

DVD video does not allow 60p footage. If you want 60i, then no deinterlacing should be done at all (at any point in the workflow). If you do want to deinterlace the footage (and make a DVD using 30p footage), with 1080 line video, you have enough lines to be able to simply drop a field (no further action required to deinterlace the footage) and yet have enough lines in each frame left to still be downsizing (result has a full 480 lines of resolution).

Originally Posted by Randall Leong

And yes, in my earlier days (when I first used AVISynth/VirtualDub), I used to drop one field then deinterlace the rest--but the videos I got were all a flickering, choppy, stuttering mess. I want smooth motion, not this stuttering flickering cr-p, when my original source is native 60i video. Dropping a field only works correctly with 30p video that had been encoded within a 60i container, IMHO.

Of course dropping a field and then "deinterlacing" will result in a mess. Dropping a field deinterlaces the footage, in and of itself. Every line in each individual field is acquired at the same time. The individual fields are progressive images by nature. If you drop one of the fields from 60i footage, what you are left with is defacto 30p footage. Applying a deinterlacing filter, at that point, is completely nonsensical.

Originally Posted by Randall Leong

And you were still thinking that 60i equals 30p. That is incorrect. The "shutter" in a 60i camera refreshes every 1/60 second--and each 1/60 second contains a different image. A native 30p camera, on the other hand, either refreshes only every 1/30 second or electronically duplicates every single frame to match a 60 fps frame rate.

No. Not at all. I am in no way confusing 60i for 30p source footage. What I said was, if you downsize a 1080i60 frame (shot at 1/60 shutter) to 480 lines, without telling the resizing algorithm that the source footage is interlaced (2 individual fields per frame), doing so will effectively deinterlace by blending the two fields together, and the end result will be very similar to looking the same as if the source was shot as 30p at a shutter speed of 1/30th of a second and resized to 480 lines (entirely different sources yet a very similar end result when resized that way, to SD resolution).

[Randall Leong]

Originally Posted by Robert M Wright

I have no idea why you are separating the fields, resizing, and then recombining them. If you want to resize interlaced footage, and get interlaced footage as a result, simply check the box next to "Interlaced" in the dialog for resizing and, unless I'm utterly mistaken, the resizing algorithm you choose will be applied to the fields individually (no need for all the extra fooling around).

From what I read, the built-in "Interlaced" setting within the resize filter only works if the final output is to be PAL, and does not work properly if the final output is to be NTSC SD. But that might have been changed since that advice was first posted.

Originally Posted by Robert M Wright

DVD video does not allow 60p footage. If you want 60i, then no deinterlacing should be done at all (at any point in the workflow). If you do want to deinterlace the footage (and make a DVD using 30p footage), with 1080 line video, you have enough lines to be able to simply drop a field (no further action required to deinterlace the footage) and yet have enough lines in each frame left to still be downsizing (result has a full 480 lines of resolution).

I know that DVD video does not allow 60p footage without a conversion to one of the supported frame rates. And I also know that already progressive footage does not need deinterlacing; in fact, deinterlacing already progressive footage actually degrades image quality.

Originally Posted by Robert M Wright

Of course dropping a field and then "deinterlacing" will result in a mess. Dropping a field deinterlaces the footage, in and of itself. Every line in each individual field is acquired at the same time. The individual fields are progressive images by nature. If you drop one of the fields from 60i footage, what you are left with is defacto 30p footage. Applying a deinterlacing filter, at that point, is completely nonsensical.

Actually, I got that huge mess even without "deinterlacing" that. This is because none of my TV sets (LCD or CRT) can properly display 30p to begin with--they handle only 60i or 60p footage correctly, and they all interpret 30p footage incorrectly.

[Randall Leong]

Originally Posted by Randall Leong

From what I read, the built-in "Interlaced" setting within the resize filter only works if the final output is to be PAL, and does not work properly if the final output is to be NTSC SD. But that might have been changed since that advice was first posted.

I have just re-checked that again, and that advice ("check 'Interlaced' in the Resize filter for PAL only; NTSC footage should be IVTC'd instead") applies only to 25p (or 24p) progressive footage that's encoded in an interlaced stream. Resizing native interlaced footage should always have "Interlaced" checked in the Resize filter.

And I am learning to get to a lot of the "simpler" features that I would end up using most often in these freeware programs. After all, my brother always told me that I have been making things harder than they have to.

Thus, my "standard" workflow to convert 1080i footage to 480i for DVD will be to apply no more than two filters in VirtualDub (and use AVISynth only to convert the color space to one which the VirtualDub filters support): Resize (with the "Interlaced" box checked) and Field Delay (to convert TFF/UFF to BFF/LFF). But I need to convert TFF to BFF only if the resulting DVD is to be played back on an older interlaced-output-only NTSC DVD player which may flicker badly when playing back TFF-interlaced footage or if I want to hand over copies of my downconverted SD DVDs to family or friends (since I would never be certain just how old the DVD playback equipment they might have) to ensure compatibility. The TFF to BFF conversion is not necessary if I were making personal DVDs for myself that are to be played back on a modern progressive-scan or upconverting DVD player.

If the footage to be downconverted to 480i SD is from a 720p broadcast transmission whose original video source had been interlaced, then I leave the "Interlaced" box unchecked in the Resize filter but add the Interlace filter after Resize but before Field Delay. In the Interlace filter, I select Progressive Frames in the source and either Even Field First (this makes a TFF interlaced video) or Odd Field First (for BFF interlaced result), depending on the field dominance of the original source video, in the field order selection. And if the field dominance does not quite match what I expected (for example, the original video's top field falls into the broadcast's odd field as detected by VirtualDub), then I apply the Field Swap filter after the Interlace filter.

[Randall Leong]

Originally Posted by Randall Leong

Thus, my "standard" workflow to convert 1080i footage to 480i for DVD will be to apply no more than two filters in VirtualDub (and use AVISynth only to convert the color space to one which the VirtualDub filters support): Resize (with the "Interlaced" box checked) and Field Delay (to convert TFF/UFF to BFF/LFF). But I need to convert TFF to BFF only if the resulting DVD is to be played back on an older interlaced-output-only NTSC DVD player which may flicker badly when playing back TFF-interlaced footage or if I want to hand over copies of my downconverted SD DVDs to family or friends (since I would never be certain just how old the DVD playback equipment they might have) to ensure compatibility. The TFF to BFF conversion is not necessary if I were making personal DVDs for myself that are to be played back on a modern progressive-scan or upconverting DVD player.

It turned out with this experiment that my video still came out somewhat jittery. I was using one of the wrong filters in this workflow.

After further experimenting, the correct 1080i TFF to 480i BFF workflow in VirtualDub is the Resize (with the "Interlaced" box checked) filter, then the Field Swap (not Field Delay) filter applied after the resizing. (Or, if I omit the Field Swap filter in VirtualDub, I would need to create an AVISynth script containing at least the line "ComplementParity" in the .avs file.) I noticed this when I separated the fields in the new file rendered with my previous workflow, and the wrong field got presented first after stepping through part of the video. The Field Swap filter takes care of the "wrong field displayed first" problem; it does not convert BFF to TFF or vice versa (as I have recently discovered). This new (to me) workflow creates the proper 480i BFF downconversion that's needed for NTSC standard-definition. No tweaks other than the conversion of the color space back to YV12 (4:2:0) are required.

Converting a 720p stream created from originally interlaced video to 480i requires another workflow that I have not yet attempted. I will experiment with the VirtualDub settings for that workflow sometime in the near future. But for now, I am optimizing my 1080i to NTSC 480i workflow with several videos.

[Pavel Tomanec]

Originally Posted by Steve Rotter

thanks Pavel, i will do that. will panning be even more jittery if shooting in 24p? i like the smoothness of 60i but it's too clean...like video cameras from the 90s.

Hi Steve,

You see, if you wish to get a bit filmic feel for your work consider not pan and zoom too much, zooming should be avoided in 95% situations. If you pan use true fluid head for your tripod or gentle movement when using steadycam. You can shoot your pans in 60i and then de-interlace on the timeline using Natress or other software packages. You can add a bit of motion blur too to soften the feel of the movement.

If you wish to come closer to perfection consider getting 35mm adapter and manual lenses, focus puling and high quality ND graduated filter and couple of diffusers and good reflectors for low budged work.

Oops, forgot to mention quality mikes for indoor and outdoor!

Hope this helps.

[Pavel Tomanec]

Hi All,

I'd like to add that I achieved relatively very good quality result using Compressor with frame controls fro HD to PAL. In my view the final DVD was much better then the output from TMPGEnc 2.5 using best resizing filter.

Pavel

[Randall Leong]

Originally Posted by Randall Leong

It turned out with this experiment that my video still came out somewhat jittery. I was using one of the wrong filters in this workflow.

After further experimenting, the correct 1080i TFF to 480i BFF workflow in VirtualDub is the Resize (with the "Interlaced" box checked) filter, then the Field Swap (not Field Delay) filter applied after the resizing. (Or, if I omit the Field Swap filter in VirtualDub, I would need to create an AVISynth script containing at least the line "ComplementParity" in the .avs file.) I noticed this when I separated the fields in the new file rendered with my previous workflow, and the wrong field got presented first after stepping through part of the video. The Field Swap filter takes care of the "wrong field displayed first" problem; it does not convert BFF to TFF or vice versa (as I have recently discovered). This new (to me) workflow creates the proper 480i BFF downconversion that's needed for NTSC standard-definition. No tweaks other than the conversion of the color space back to YV12 (4:2:0) are required.

Converting a 720p stream created from originally interlaced video to 480i requires another workflow that I have not yet attempted. I will experiment with the VirtualDub settings for that workflow sometime in the near future. But for now, I am optimizing my 1080i to NTSC 480i workflow with several videos.

Another boo-boo in this workflow, as it turned out. The Field Swap filter got the field order wrong again since AVISynth itself got the field order wrong on AVI footage. The problem is that on AVI files imported through AVISynth, VirtualDub reported reversed field dominance. Unless I specify the field dominance ("AssumeTFF" or "AssumeBFF") or use "ComplementParity" in AVISynth, VirtualDub will assume all interlaced AVI imports from AVISynth as BFF. This incorrect reporting of reversed field dominance did not affect the final output.

So, I have reverted to simply applying the Resize filter with "Interlaced" checked within the filter. The field dominance of the newly downconverted file, in this case, remains the same as the original HD interlaced footage (TFF/UFF). Modern NTSC DVD players will automatically convert any TFF footage to BFF before outputting through the S-Video or composite video outs.