Blender

Durian DVD

Follow our progress!

  • on Twitter
  • on Identi.ca
  • Follow our commits on Twitter!

    Sintel 4k, 3×16 bits available too

    on February 25th, 2011, by Ton

    After almost 10 days of uploading, the 650 GB of 16 bits TIF files arrived safely at xiph.org. Ralph from Xiph converted these to 16 bits PNG and mentioned it was actually 165 GB less data 😉 Both TIF and PNG are available there now.

    http://media.xiph.org/sintel/

    Hopefully they can manage the bandwidth. Thanks for hosting it xiph.org!

    We’re very interested to host and link to brand new superior encoded HD Sintel versions. Post urls here!

    36 Responses to “Sintel 4k, 3×16 bits available too”

    1. Teddl Says:

      Not superior and
      I’ve posted this before but maybe it was catched by anti-spam?
      Originaly posted at the bay..
      Now to be save, just google for
      Sintel anti banding

    2. Gianka Says:

      Why not using webm format? Is better than H264 and OpenSource!!!

    3. Franck Says:

      Good work ! Will other stuffs like concepts, scripts, storyboards or blends be published online ?

      @Gianka : never tested myself, but it is often said that WebM is better than Ogg, however is still worse than H264. Anyway, Sintel is actually a great “material” to do tests on. 🙂 (Sorry for my English)

      For Durianners who liked/loved Sintel, you should look at the new amazing trailer for the game Elder Scrolls : Skyrim –> http://www.youtube.com/watch?v=ArbzAnHGpks&hd=1
      A quite strange thing, the design of the dragon reminds me David’s one, with the “wing-arm” and the horns. ^^ But I think that it is a coincidence, they began to work on it 5 years ago.

    4. Teddl Says:

      16bit pngs is finaly good news for those who see a different between them and the 8bit ones. I don’t, at least not yet. No screening in my area yet.. :p

      @Gianka: Good question! That’s what Google uses on Youtube right?

      However, if someone wants to redo antibanding or can improve it (maybe from the 16bit pngs), here is how I did it:
      http://i652.photobucket.com/albums/uu244/Teddl1/Howto2.png

    5. Mathias Says:

      If you don’t see a difference between them, look at this comparison:
      http://screenshotcomparison.com/comparison/31327
      Its from a dithered encode I did of the opening fade:
      http://www.multiupload.com/LED0PVYUWD

      I will upload dithered encodes in 720p and 1080p when I got the complete source.

    6. Teddl Says:

      @Mathias
      I definitely applied a stronger filter that was more pleasant to my eyes.

      Will download your version too!
      -Teddl

    7. Mathias Says:

      @Teddl The workflow for the screenshot comparison was:
      1) Convert to Double precision YUV 4:2:0 (so it only had quarter the resolution in U and V planes)
      2) Dither to YV12 (or don’t dither for the non-dithered version)
      3) Encode at 4Mbit/s
      4) Take screenshots

      So it only has quarter res color planes and already was encoded, which minimizes the dithering effect.
      I’m only on my laptop right now, so I can’t judge the quality properly, but it still looks better to me.
      If you have any suggestions to the process, feel free to give me suggestions.
      Also I’m also interested in your workflow, what software you use and similar. I found very few software that was able to properly dither 16-bit to 8-bit (and preferably in YUV 4:2:0).

    8. Teddl Says:

      @Mathias
      Yours looks better, I agree. I assume you keep uncompressed sources until the final encode. That’s a big plus. Because of low diskspace I had to use HuffYuv. However, now I see the “color fades” where the banding occured.
      I wanted to totaly get rid of it and my “workflow” is in the howto.png I posted earlier.

      About dithering in 16-bit:
      Have you tried to put the images into Blender’s Videoeditor? Use the dither in render options and from there into 4:2:0?
      Otherwise “Avisynth”, “Dither 1.5” and “VirtualDub” may be worth a look.
      Fits perfectly if you aren’t allready using it: “Error diffusion filter for Virtualdub”. I just wasn’t patient enough to learn how it works. The “Anti-Banding option in Donald Graft’s filter realy came handy and did exactly what I wanted.

      Suggestions:
      As I said, I see banding in the 16bit pngs too. So why not dither there (also)? Or is it my LCD Display?
      Encoding reduces dithering at a certain bitrate. Higher Bitrate or playing in x264 with –tune film or manualy setting the deblocking to 0:-1 or more or even psy-rd values can help too.

      Can’t wait for yours. Since it’s from 16bit source. 😀
      And at last, my encoded 1280×544 version can be found via Google. Search for
      Sintel anti banding 1280

      P.S.: Thanks for spreading the 1920 version so fast! You guys rock!

    9. Mathias Says:

      As far as I know, Avisynth can only work with 8-bit material. The same goes for Virtualdub with the additional drawback, that Vdub can only work in RGB colorspace.
      I didn’t try Blender. But since I don’t think it works in YV12 and Ton told me it only uses random dithering instead of any kind of error diffusion it will be inferior to my solution.

      I didn’t manage to get proper results from Imagemagick and I looked into the libavfilter sources and I think the method they use to work with 16-bit material is just to drop the lower 8 bits.
      Also when you view 16-bit sources on your monitor, chances are >99% your monitor uses only 8-bit. And since I didn’t find a single program that properly dithers to 8-bit, your viewer will also just drop the lower bits. Thats why you also see banding with the 16-bit sources.
      There is no need to dither in 16-bit, because you have no higher-bit source to dither from.
      Of course you could use debanding filters, but if you have the 16-bit sources, dithering is the way to go.

      After all, I wrote my own code to read PNGs and dither to raw YV12 to get the maximum quality in the final colorspace. x264 can directly read that format, thus the maximum quality should be retained. I have a double precision workflow until the final dithering to yv12.

      If there is any interest, I could also losslessly compress my 8-bit dithered version and upload it, then everyone could make their own encodes from a high quality 8-bit source.

    10. Riboshom Says:

      Screw tron, frame 00014478.png is my new wallpaper!

    11. Teddl Says:

      You’re right, Mathias. Avisynth doesn’t support 16-bit for now and VirtualDub works the 16-bit RGB. Support could come with Avisynth 2.7 if somebody figures it out. And yes, when I bought the monitor, I didn’t even know there is something called color depth.
      If you sharpen the video before dithering after resizing, I trust you can encode it very well. I’ve made bad experiments using sharpening after debanding.

      There is just 1 question left: Could your dithering-code somehow be implemented in Blender? I’ve heared rumors of two new Blender Institute Projects where it may come handy I presume. 😉 Just thinking.

    12. Teddl Says:

      Just to clarify what I meant with pleasant to my eye:
      http://screenshotcomparison.com/comparison/31953
      Proper dithering is from Mathias and strong anti banding is from me.
      Both a real alternative to just adding non Uniform Noise Luma 11 and Chroma 10 with ffdshow. You see that on DVDs and BluRays – looks only good on screenshots if you ask me.

    13. joeri Says:

      Dithering has less banding but more digital noise.
      Tiny flickering blixs. Don’t you just hate the digital world?

    14. Mathias Says:

      Blender already has some support for dithering, but only by randomly dithering and not error diffusion dithering.
      Dithering is not so important for blender, when you can create 16-bit files that don’t need dithering, its only when you view it on 8-bit monitors (or use 8-bit encoding) that you need dithering.
      Of course it would be nice to have the option to export a dithered version. But even then when you want to encode it, its probably (very slightly) better to dither in the final yv12 colorspace.

      And to the comparison, Teddl keep in mind that the screenshot from me already had a lossy compression applied and thus dithering was removed.
      Here is a comparison with a lossless compression:
      http://screenshotcomparison.com/comparison/32237
      IMO there is next to no banding left and I also don’t see much noise in the image.

      By the way, I have around 50% of the 16-bit frames now. After my weekend download was stopped after 15GB, I’m now downloading them at full speed again and expect to have them by Friday morning.

    15. Teddl Says:

      Mathias, do you decode the video with ffdshow?

    16. Mathias Says:

      You mean my encode? Where my comparison PNGs come from?
      I decode it with mplayer. I use the following simple command line:
      mplayer test.mkv -vo png

      Why are you asking? Is there a color difference you are seeing?

    17. Chris Eller Says:

      I’ll start pulling down the 16 bit frames now. The DCP specification supports 12 bit color, we’ll see how that looks compared to the 8 bit version. I’ll post the 16 bit DCP on our ftp server when all is said and done.

      Chris

    18. androvsky Says:

      I found a batch of bad frames in the 4k 16-bit png set…

      00011236.png
      00011273.png
      00011351.png ?
      00011416.png ?
      00011480.png
      00011497.png
      00011573.png
      00011590.png
      00011598.png

      The ones with question marks are notably darker than the surrounding frames; they’re probably broken but not as obviously as the others. Keep in mind I’m downloading out of order to get the best scenes for tech demos first, so I can’t say these are the only ones yet. The scene after this one looks a lot better so far…

      By the way, thanks again to Sintel for providing these frames and xiph.org for hosting them. 🙂

    19. Ton Says:

      Fixed frames can be downloaded here:
      http://download.blender.org/durian/Sintel_tiff16_update_march9.zip

      Thanks for the quality checks!

    20. androvsky Says:

      It’s the least I can do, thanks for the quick update 🙂

    21. Ralph Giles Says:

      I’ve updated media.xiph.org, except for 00011666 which is still too dark.

      You might also want to try our experimental rsync support. You can now say,

      rsync rsync://media.xiph.org/

      and e.g.

      rsync -av rsync://media.xiph.org/sintel/sintel-1080-png .

      This is very helpful for resuming downloads or updating your file set after fixes like this. You can also grab the latest copy of the MD5SUMS.txt file in each directory an compare against your download to see what’s changed.

      I’ve posted a 1k 16-bit png fileset derived from the 4k set, which might be helpful to those wanting to play with 16 bit sources without having to download half a TB,

    22. Ton Says:

      Another update on the 16 bits sequence has happened. It appeared file copy (rsync) of USB drives is not that reliable as I hoped…

    23. Robbert Says:

      I wonder were there (named) chapters on the dvd for Sintel?

      If so would somebody post the time codes (and names) so we can use them to put keyframes at the right places for example an Blu-Ray/AVCHD encode.

    24. Mathias Says:

      Frame 00011666 is still broken in the repository, even though it was replaced during the process.
      Its much darker then the surrounding frames.
      I know now I’m nitpicking, but why is Vivian Haber’s name underlined in Frame 00020873?
      Would be awesome, if at least the first problem could be fixed.

    25. Ralph Giles Says:

      Frame 00011666 is updated now.

    26. Mathias Says:

      Finally, I didn’t see any more broken frames. Awesome!
      @Robbert: There are chapters on the DVD, but they are not named. This is what were on the PAL DVD (the NTSC one was kinda funky because of the 29.97 fps).
      “””
      2432
      1104
      4858
      2084
      5811
      1580
      3387
      435
      “””
      Those are the length in frames. However the new source starts 3 frames late (and the very last shot is one frame longer, also the credits are a bit different). So if I factor that and convert it to timecodes at 24 fps you get
      “””
      00:00:00.000
      00:01:41.458
      00:02:27.583
      00:05:50.125
      00:07:17.083
      00:11:19.333
      00:12:25.292
      00:14:46.583
      “””
      However, I think some chapters should be changed by a few frames and there is no chapter for Sintel entering the Dungeon, but one for the CC license at the end… I changed them a bit and this is what I use now:
      “””
      00:00:00.000
      00:01:43.125
      00:02:28.667
      00:05:49.792
      00:07:17.208
      00:07:52.75
      00:11:18.833
      00:12:24.083
      “””

    27. Mathias Says:

      I just realized: The DivX encode has named chapters:

      Menu
      00:00:11.167 : en:Prologue – fr:Prologue – de:Prolog – it:Prologo – pl:Prolog – ru:Пролог – es:Prólogo
      00:00:40.417 : en:Snow Field – fr:Neige sur le terrain – de:Schneefeld – it:Nevaio – pl:Pole śniegu – ru:снежном поле – es:campo de nieve
      00:00:51.875 : en:Attack! – fr:Attaque! – de:Angriff! – it:Attacco! – pl:Atak! – ru:Hападение! – es:Ataque!
      00:01:52.167 : en:Rescued – fr:Sauvé – de:Gerettet – it:Salvato – pl:Uratowany – ru:Cпасенный – es:Rescatados
      00:02:42.625 : en:Town – fr:Ville – de:Stadt – it:Città – pl:Miasto – ru:город – es:Ciudad
      00:03:34.459 : en:New friend – fr:Nouvel Ami – de:Neuen Freund – it:Nuovo Amico – pl:Nowy Przyjaciel – ru:новый друг – es:Nuevo Amigo
      00:06:00.000 : en:Departure – fr:Départ – de:Abreise – it:Partenza – pl:Wyjścia – ru:Bыезд – es:De Salida
      00:07:27.667 : en:Back In Hut – fr:Retour dans la cabane – de:Zurück in der Hütte – it:Torna en Capanna – pl:Z Powrotem w Chacie – ru:Oбратно в Xижину – es:De Nuevo en la Choza
      00:08:03.000 : en:Cave – fr:Grotte – de:Höhle – it:Grotta – pl:Jaskinia – ru:пещера – es:Cueva
      00:11:10.000 : en:Escape – fr:échapper à – de:Flucht – it:Fuga – pl:Uciec – ru:побег – es:Escapar
      00:12:35.000 : en:Credits – fr:Crédits – de:Kredite – it:Crediti – pl:Kredyty – ru:Kредитов – es:Créditos

      But they have a logo ahead of the movie, so the timecodes are kinda different…

    28. joeri Says:

      48 Frames Per Second is going to happen.

      Jackson is shooting the Hobbit in 48 fps.
      https://fbcdn-sphotos-a.akamaihd.net/hphotos-ak-snc6/215793_10150222876561558_141884481557_8844575_4624739_n.jpg

      https://www.facebook.com/note.php?note_id=10150222861171558

    29. joeri Says:

      The dvd authoring tool puts chapters on I frames.

      Not all chapters where encoded by chapter ( lots where though for better image quality ) so sometimes the I frame is not on the correct point.

      The chapters where a last minute addition, sorry to hear the nstc version did not survive the conversion. I didn’t think entering the dungeon was a new chapter, but I did not consult Colin about the chapters due to time restrictions.

    30. Kalle Says:

      Can anyone encode it with h.264 in 4K with a healthy bitrate? Preferably with 5.1 audio. That would be great demo material for the new Ultra HD TVs that I hear soon will be available

    31. tim Says:

      I like the rsync…I got up to 00007104 of the 1080p PNG…I reformatted linux and backed up the files, but now when i rsync it doesn’t recognize that I downloaded all of those 7104 files. What do I need to do? I have a cap, I can’t afford to download the 7104 again just because rsync is acting up.

    32. Jonathan Says:

      From the 16bit Tiff’s, I’ve made a 4K SMPTE Compliant DCP in Full Container (4096×2160).
      I’ve simply padded the frames with black, in order to comply with the specs.

      The original frame (4096×1744) is just a bit too big to fit in a Scope 4K frame (4096×1716) without cropping or resizing it. So by keeping the original width and just padding top and bottom with black, all original pixels is kept 🙂

      I’ve tested this in a cinema on a Doremi DCP2000 and a Barco DCP3000 (a 2K projector). All looked and sounded fine.
      The Audio is a FLAC -> WAV transcode from the 5.1 FLAC Master. The audio is not modified in any way.

      Tools used (all open source):
      ImageMagick
      FFmpeg
      SoX
      OpenDCP

      I’m hosting it on my private fiber connection, so I’ve limited the ftp to max 5 users at a time. Hopefully that’ll give some OK speeds.
      Please bare with eventual server crashes.
      Enjoy! 🙂

      ip: bombjack.dk
      username: dcp
      password: freedcp

    33. Jonathan Says:

      Just to add to the above post.
      I’m assuming people who download the DCP (Digital Cinema Package) know what it is and know what to do with it, but for those who are curious, please see Chris Eller’s post here:
      http://www.sintel.org/news/sintel-4k-version-available/#comment-63651
      His DCP is INTEROP and created from the 8bit TIFF’s, “my” DCP is SMPTE and created from the 16bit TIFF’s. Just to clarify 🙂
      The audio seems to be higher in my DCP, compared to Chris’. I’m not sure why.
      I’ll probably do an INTEROP version sometime soon (when I have added some space to my ftp).

      //Jonathan
      Projectionist from the cold north.

    34. Mathias Says:

      @joeri I didn’t say the NTSC chapters werent working, I said I had problems extracting the correct framecount/timecodes from the NTSC version, because with NTSC you put 23.976 fps into 29.97fps video and it was easier for me to use the PAL timecodes.

    35. Jonathan Says:

      I’ve taken the above ftp down for the time being, but will make it available pr. request.

    36. Adrian Says:

      Hi, i’m downloading the hole folder of 4k 16bit sintel. I saw that the H.264 MKV video available for download has a 10Mb/s bitrate. I’ll compile/encode all the PNGs to a H264 MKV video of 30Mb/s of bitrate (higher quality) Maybe I’ll upload-it here when I finish so that everybody can download the video with a higher quiality. Sorry for my bad English lol, I’m from Madrid.
      @Kalle I’ll try to convert the PNGs to 4k H264 50Mb/s bitrate too.
      Is there something I should know before compiling/encoding the video? Thanks
      Who ever wants can contact me on [email protected]