DV to H.264 for web

[HDClown]

I'm looking to convert various DV clips. These are 8mm/Hi8 shot on consumer Sony camcorders. I imported with WinDV as type 2 AVI in DV codec over firewire.

I'm looking to deinterlace and convert so these can be uploaded to the web. Various sites like Facebook, YouTube, Vimeo, SmugMug.

I've been trying to find something that is fairly easy to setup, but everything I find has a lot of settings to figure out. As of now, I've determined that I shoulde interlace with AviSynth and YadifMod()+NEEDI2 is a good combination for quality and speed. After that, it seems I should ConvertToYv12(). The problem with portion is I don't know what flags to use.

After this step, I'm not really sure what to do. I played with MeGUI but it looks like it wants separate audio and video input, so I'd need to demux, not sure best way to do that. I'm not sure what settings I should be using in MeGUI for the encoding, given the intended output for web. Is there something easier than MeGUI? Does someone have some good recommendations on settings to use for the web?

File details:

Video
ID : 0
Format : Digital Video
Codec ID : dvsd
Codec ID/Hint : Sony
Duration : 8mn 22s
Bit rate mode : Constant
Bit rate : 24.4 Mbps
Width : 720 pixels
Height : 480 pixels
Display aspect ratio : 3:2
Original display aspect ratio : 4:3
Frame rate mode : Constant
Frame rate : 29.970 fps
Standard : NTSC
Resolution : 8 bits
Colorimetry : 4:1:1
Scan type : Interlaced
Bits/(Pixel*Frame) : 2.360
Stream size : 1.68 GiB (95%)

Audio
Format : PCM
Format settings, Endianness : Little
Format settings, Sign : Unsigned
Codec ID : 1
Codec ID/Hint : Microsoft
Duration : 8mn 22s
Bit rate mode : Constant
Bit rate : 1 536 Kbps
Channel(s) : 2 channels
Sampling rate : 48.0 KHz
Resolution : 16 bits
Stream size : 91.9 MiB (5%)
Interleave, duration : 33 ms (1.00 video frame)
Interleave, preload duration : 500 ms

[edDV]

Is your goal ultimate quality?

Describe the video content focusing on motion and lighting. The more the action or the higher the noise caused by poor lighting, the more time you will need in preparation. The main downside of deinterlace is loss of half the motion samples.

If the video is relatively static, and/or you can live with shortcut artifacts, simplified programs exist like Handbreak offer a shortcut to 29.97p h.264 that listed services will accept.

You might do a test upload to observe the simple solution.

[HDClown]

Is your goal ultimate quality?

Describe the video content focusing on motion and lighting. The more the action or the higher the noise caused by poor lighting, the more time you will need in preparation. The main downside of deinterlace is loss of half the motion samples.

If the video is relatively static, and/or you can live with shortcut artifacts, simplified programs exist like Handbreak offer a shortcut to 29.97p h.264 that listed services will accept.

You might do a test upload to observe the simple solution.

The content will vary greatly in it's setting, but the general content I am working with now is the same, it's marching bands/drumline. Content can be close up or zoomed out of the entire field. Daytime or night time. There's always lots of moving going on regardless. The source cameras and material are of low-average to average quality in general (compared to modern camcorders). This is stuff recorded 10 years ago (to give you an idea of age of the camcorders used to record it).

I'm not looking to improve the quality of the source material right now, I just don't have a fast enough CPU (and the associated time) to get into that. I've seen some of the impressive things AviSynth can do in this regard, but that's a future project. Right now I would just like to deinterlace and encode to H.264 without entirely killing the original quality. Most of the places I would upload to (Facebook, YouTube in particular) are going to re-encode the video a 2nd time anyway.

I did a test just now in Handbrake using the preset of "High profile" as a start. i set the video for average bitrate of 1100kps and used the decomb-standard setting. This seemed to meet my goal of not butchering original video while giving me a much smaller H.264 file. So I guess the question becomes, is there any reason to get into a more involved/complicated process right now?

Lastly, I've noticed is that the output video at "Actual size" is 720x480, but the source video shows at "Actual size" at 640x480. I actually saw this when playing with MeGUI and exporting form QuickTime Pro as well. I set the encoders to use the resolution of the source but the output file sizes larger than the source. What's going on with that?

[edDV]

I did a test just now in Handbrake using the preset of "High profile" as a start. i set the video for average bitrate of 1100kps and used the decomb-standard setting. This seemed to meet my goal of not butchering original video while giving me a much smaller H.264 file. So I guess the question becomes, is there any reason to get into a more involved/complicated process right now?

That is why I suggested you test the easy path and then figure out what further problems need solution.

Lastly, I've noticed is that the output video at "Actual size" is 720x480, but the source video shows at "Actual size" at 640x480. I actually saw this when playing with MeGUI and exporting form QuickTime Pro as well. I set the encoders to use the resolution of the source but the output file sizes larger than the source. What's going on with that?

Much of the AVC world is still obsessed with square pixels and yes I agree you need to make these compromises. If your video was 16:9, they would have forced you to attack vertical resolution.

Never assume these compromises are reversible.

[edDV]

My computer crashed from summer heat immediately after I typed that. Yes it was busy encoding 100%.

When evaluating quality, remember the original is 25 Mb/s for video and your result is 1.1 Mb/s including audio.

I sit in amazement. Still I keep the full bit rate as archive. Don't assume what "looks good" on a PC today will look good in twenty years.

[HDClown]

These H.264 encodes are strictly to put out on the web. The original imports from DV stay untouched as masters Since the online streaming sites all re-encode your content anyway, I'm partially at their mercy. There are also file size limitations on how big of a file they accept, so that has to be factored in. YouTube is 1gig/10 min max, Facebook is 1gig/20 min max.

I started with 1.1 Mb/s to see what the overall size was like, since I have to upload these files at only 1Mb/s. My 8 meg clip ended up about 80 megs, so I can up the bitrate considerably. I'll probably go into the 4-5Mb/s range. That seems to be a common range preset in a few different encoders I've played around with, and it would confine within the limits of places I need to upload to.

In regards to the resolution size, is there a generally recommended option?

[creamyhorror]

HDClown, if Handbrake produces acceptable quality to your eyes, then you can just go with it. Avisynth's main advantage in this case is that it has potentially better deinterlacers - Yadifmod+NNEDI2, as you noted, and TempGaussMC. But they may not be worth your time to investigate unless you're going for top quality.

In Handbrake, High Profile is a good base to start with. I don't think Youtube deals properly with anamorphic video, so you have to resize to a 4:3 ratio (resulting in square pixels as edDV said). In Handbrake, switch to "anamorphic none" under Picture settings. 640x480 results in a loss of horizontal resolution, while 720x540 preserves full horizontal resolution but upscales the vertical (requiring more bitrate). The latter is probably the better option, but I'm not sure if Handbrake allows you to upscale.

If you find you're still getting small areas of interlacing/combing in your Handbrake encode, turn Decomb off and set Deinterlace to slower. This will tell Handbrake to completely deinterlace the source instead of attempting to preserve uncombed areas.

For bitrate, you want to keep the source at a good bitrate so the re-encode done by the video-sharing site won't be as ugly. You can use an RF value of between 17 and 19 to let x264 select the bitrate based on the source's complexity, or try bitrates like 4-5Mbps.

[HDClown]

What is your suggestion in regard to detelecine? This goes to "Default" with high profile preset.

It will not let me upscale in handbrake beyond 480.

If you could check my thought here: The actual resolution of the original recorded content is 640x480 in a 4:3 format, but when I imported with WinDV, it put it into 720x480 container with display ratio of 3:2?

When I open the program in WMP and QuickTime, it's playing in 4:3 format, but in VLC, it plays in 3:2 format. I assume this has to do with internal flags in the container as to how the player renders, correct?

Now, if the original recorded content is 640x480 than wouldn't any kind of playback in a widescreen format cause distortion of the original picture, or does that to depend on how the encoder is configured? I don't want to distort the image, so I assume I can anamorphic none and force 640x480 and I'll get an encode at the original resolution? Is there a way to get a widescreen format without causing image distortion?

On the handbrake RF setting, is higher # better quality/higher bitrate, or the other way? I used 20 and it came out in the 6Mbps range.

[creamyhorror]

What is your suggestion in regard to detelecine? This goes to "Default" with high profile preset.

You can turn it off, although it's supposed to be safe to leave on. DV is interlaced so the Deinterlace or Decomb option is what you want (not both at the same time though).

It will not let me upscale in handbrake beyond 480.

Well, no helping that then.

If you could check my thought here: The actual resolution of the original recorded content is 640x480 in a 4:3 format, but when I imported with WinDV, it put it into 720x480 container with display ratio of 3:2?

The actual resolution is 720x480, but the proper display aspect ratio should be 4:3 (fullscreen), if I read understand your MediaInfo output right.

When I open the program in WMP and QuickTime, it's playing in 4:3 format, but in VLC, it plays in 3:2 format. I assume this has to do with internal flags in the container as to how the player renders, correct?

Yes, but it should be consistent between different players. Maybe WinDV is doing something funky with the display aspect ratio flag, I dunno.

Now, if the original recorded content is 640x480 than wouldn't any kind of playback in a widescreen format cause distortion of the original picture, or does that to depend on how the encoder is configured? I don't want to distort the image, so I assume I can anamorphic none and force 640x480 and I'll get an encode at the original resolution?

You'll get an encode at the correct display aspect ratio (4:3, since 640:480 = 4:3), not at the original resolution (which is 720x480).

Is there a way to get a widescreen format without causing image distortion?

No, since the correct display aspect ratio is 4:3. Unless you mean something else?

On the handbrake RF setting, is higher # better quality/higher bitrate, or the other way? I used 20 and it came out in the 6Mbps range.

The lower the RF, the higher the quality or bitrate. I guess this source is a difficult one. You can increase the RF if you find the bitrate too high.

[HDClown]

Looking at some specs, Video8 is "half-NTSC" resolution and Hi8 is "full NTSC resolution". This equates to 240 or 480 in modern digital standards we are used to. Digital8 combines Hi8 with DV codec to encode the format digitally on the cassette. So the actual resolution is closest to 640x480 it seems, and that by nature of the camera I am using to import using DV codec, it gets imported in a way and flagged that it's 720x480. It's interesting to see in QuickTimeX on my Mac how is whos as "720x480 (640x480)". It's reading some different data that media info doesn't even pick out. But WMP also chooses to display in 640x480, even through the container says resolution is 720x480. It must be because the original aspect ration is encoded as 4:3 but the display aspect ratio is 3:2 per the 720x480 resolution. VLC doesn't want to honor original ratio, but WMP/QT do.

As far as I know, WinDV doesn't do anything but pull data off the device as a raw stream, so it's not messing with the digital format.

Ultimately, to make sure the encoded file is matched in resolution and aspect to the original recording in-camera resolution/aspect, it seems like I should do anamorphic none and set 640x480 (which requires unchecking maintain aspect ratio).

[creamyhorror]

Ultimately, to make sure the encoded file is matched in resolution and aspect to the original recording in-camera resolution/aspect, it seems like I should do anamorphic none and set 640x480 (which requires unchecking maintain aspect ratio).

That should be right. I don't have experience with DV so I can't comment on the rest of what you wrote.