MPEG4 H.264 encoding tests (vs Xvid)

13»

Comments

  • Reply 41 of 59
    ebbyebby Posts: 3,110member
    I encoded the Incredibles Yesterday. Using H.264, I was getting roughly 8FPS and it took all day. But using MPEG-4 (FFMPEG) I was getting about 90FPS (First pass) and 45FPS (Second pass). I'll try encoding H.264 using FFMPEG later, but I have class tonight.
  • Reply 42 of 59
    If anyone wants further clarification that H.264 is superb, I have Minority Report screenshot comparisons.



    I set the bitrate to 700Mbps and the resolution to 720 x 304.



    1) H.264 encoding, 2-pass, 128Kbps AAC Audio, MP4 file - Came out at 826.9MB.



    2) Xvid encoding (FFMpeg), 2-pass, 128Kbps AAC Audio, MP4 file. - Came out at 843.6MB.



    The H.264 encoded movie (if you need it pointing out ) is the AVC window, the Xvid being the QT window.



    Picture 1

    Picture 2

    Picture 3



    There's no apparently slow down when it's playing, but if I try to skip through, VLC will give me a 5 sec no video warning, basically due to a slow computer - 1.33Ghz, 512MB RAM. I have some Futurama episodes, encoded at the same as the above H.264, and playing slows during the opening credits. I hope (wish) QT's playing will be a bit more streamlined so that this doesn't happen. \
  • Reply 43 of 59
    123123 Posts: 278member
    Quote:

    Originally posted by danielctull

    The H.264 encoded movie (if you need it pointing out ) is the AVC window, the Xvid being the QT window.



    really? The QT movie looks much better.



    BTW, use the same app for both movies (same post processing).
  • Reply 44 of 59
    Indeed, the image looks sharper in quicktime.



    I noticed that reading an H.264 video requires much more power than decoding an Xvid (or divx, etc). And as we said before, the encoding is MUCH longer with H.264 (3 times longer with handbrake, 6 times in quicktime 7).

    The image quality may be better, but these are real tradeoffs, especially for people not having a high-end computer. \



    This is from MOSR :

    Quote:

    No personal computer on the market today -- not even the most powerful, most expensive dual-processor Intel Xeon or Apple G5 -- is really up to the task of processing H.264 video efficiently.



    Decoding high-quality H.264 video can be done acceptably on most Macs which support OS X, (preferably a G4 or better to avoid stuttering) but it will demand a very high percentage of the system's CPU resources - nearly 100% on anything short of a G5.



    Even on the latest dual 2.5GHz G5, decoding H.264 video demands a much larger portion of the system's resources than playback of video encoded with any other popular codec.

    \t





    Encoding is another matter entirely; high-resolution HD video encoded with H.264 is a tremendous strain even on that dual G5. Don't expect smooth multi-tasking performance while doing so -- not even under Tiger.



  • Reply 45 of 59
    kickahakickaha Posts: 8,760member
    Gotta disagree guys, the QT image is *lighter*, but it's not sharper. Zoom in, and the differences become quite evident. Take the first one, for instance. The QT window is much more pixellated, and a lot of the detail in the brighter areas (the door in the background) is just washed out completely. Zoom in on a darker area (his vest) and the H.264 window still shows level of detail, while the other is just chunky.



    I can deal with a bit of brightness balance for that level of increased clarity.
  • Reply 46 of 59
    Quote:

    Originally posted by ZO

    I just realized something though..



    I dont get why the XVID file is bigger eventhough it says its a lower bitrate. That makes no sense. It should be even smaller.



    Weird.




    I don't know exactly how H.264 encodes videos, but I assume it's similar to XVID or DIVX. In DIVX, what you set is actually the estimated average bitrate since the actual bitrate varies at any given time. There are other values you must set: Qmin and Qmax which tell the encoder how good a quality you expect of the video vs. how big the file size ends up. You may want to check the final average bitrate in QuickTime > Movie > Get Movie Properties (CMD-J) and select Video Track and General pull down menus (reminder: it's in Kbytes/s, not kbits/s).



    Presumably, at your settings, H.264 achieves your required quality with less of your desired average bitrate while XVID must exceed your desired average bitrate. For a 2-hour video, the audio size is about 112.5MB, which leaves roughly 562.9MB (640.4 kbps) and 587.6MB (668.6 kbps) for video in H.264 and XVID. So actually it makes sense.



    Quote:

    Originally posted by danielctull

    There's no apparently slow down when it's playing, but if I try to skip through, VLC will give me a 5 sec no video warning, basically due to a slow computer - 1.33Ghz, 512MB RAM. I have some Futurama episodes, encoded at the same as the above H.264, and playing slows during the opening credits. I hope (wish) QT's playing will be a bit more streamlined so that this doesn't happen.



    That makes me wonder (and concerned) what the cut-off hardware is for playing an HD movie. I have a sub-1GHz iMac G4.
  • Reply 47 of 59
    zozo Posts: 3,117member
    playing HD is a bitch...
  • Reply 48 of 59
    hmurchisonhmurchison Posts: 12,435member
    Now people realize why we need CELL



    Even before QT7 I remember reading some codec sites that claimed that it requires 8x the processing speed to encode AVC relative to MPEG2 and 4x the speed to decode.



    You can forget iTMS movies using AVC for a few years while everyone increases the speed of their computers. Just because the files are smaller doesn't mean it won't require a lot of juice to play them back.



    This of course is assuming we're talking about HD resolution. SD rez shouldn't be quite that bad.
  • Reply 49 of 59
    Unless I'm missing something... the window to the lower-right in each of those screengrabs was MUCH better in each instance.



    There was a greater tonal range and clearer details. This is true for each of those shots... are you sure you don't have it backwards...?



    Look at the shot with Colin and Tom speaking... Tom Cruise's face has MUCH more detail in the lower-right window... in the tub scene... look at the mid-range values... much clearer in the Btm-Rght one and look at the washcloth on the edge of the tub... much sharper detail in the Lwr-Rght one.



    Again... am I missing something...?
  • Reply 50 of 59
    kickahakickaha Posts: 8,760member
    Take Pic 3. Zoom in on his fingers of the hand in the upper portion of the picture.



    In the bottom right video, it shows a *lot* of color pixellation between his fingers and the background. Top left, almost none, it's a nice sharp defined line.



    The top left one has more detail by far, the bottom right one is using image lightening to try and make up for that. It looks 'better' until you actually zoom in and inspect the details.



    It's kind of like how some people can't stand Sony Trinitron tubes, even though they have sharper detail than most other tubes. Other tubes blend the phosphor dots causing a lightening effect, and it looks 'better' to some people's eyes, but if you look at the darker Sony picture and turn up the brightness a touch, you end up with a *much* clearer picture with a deeper color range and richer black.



    The top left image in all three screengrabs is distinctly showing more detail if you zoom in, but I'll grant that at first glance it looks too dark and smudgy.



    Also, if you consider that one of the common artifacts with video encoding is analogous to the Sharpen tool in Photoshop, et al, that can be a source of confusion. Take that washcloth. Again, if you zoom in, you can see that the boundaries on the wrinkles are 'sharp', going from dark to very light and back over just a few pixels. A soft cloth isn't going to have that sort of specular highlighting, what you're seeing is a color boundary artifact... but the effect is to make it look more crisp. It's losing detail, but by cranking up the sharpening effect a bit, and drawing your eye to more coarse features more prominently, it gives the effect of being 'crisp'... even when throwing away a lot of image information. It looks good at small sizes, but if that were blown up to say, a wall projector, it'd look really bad. The H.264 OTOH, would look much better.
  • Reply 51 of 59
    wmfwmf Posts: 1,164member
    Quote:

    Originally posted by hmurchison

    You can forget iTMS movies using AVC for a few years while everyone increases the speed of their computers. Just because the files are smaller doesn't mean it won't require a lot of juice to play them back.



    Of course, you don't have to use a Mac to watch your (hypothetical) H.264 movies downloaded from the (hypothetical) iTunes Movie Store. You could use a (hypothetical) EyeHome Extreme based on the SMP8630 chip, which would be much cheaper than putting a Mac on top of your TV.
  • Reply 52 of 59
    I honestly can't believe you guys are questioning it! To further add to Kickaha's post:



    The most obvious one there is the spyders scene in picture 3. Look at it 100% zoom and you can easily see artifacts around all of the spyders in the bottom right (xvid) movie, whilst in the top left picture there are none. The spyder in front of Tom has light reflecting off of the top. Again in the xvid movie this light is very big and blurry, whilst in the H.264 copy it is a much more defined spot of light - just how it should be.



    With pictures 1 and 2, look at the edge of their faces, lots of blurriness, especially in picture 1. Picture 1 in fact, shows how xvid seems to lighten the picture to make it look better. Look at the doors in the background and you'll notice you can see a good majority of the doorway in the H.264 copy, but in the xvid version it is very lit up, and nearly impossible to see the upper of the doors.



    I don't honestly mind about the encoding times, if it takes 3x longer hey ho, I'm getting a good copy. Playback is starting to worry me though, my PB can take only just take the playback of the H.264 movies, and sometimes skips frames or misses chunks of video. I have a HD (1280x720) episode of Alias encoded with 3ivx, and I can *just* get that to playback fine with processor on highest (not auto) with nothing else running. It's just not going to happen with H.264 that's for sure...
  • Reply 53 of 59
    hmurchisonhmurchison Posts: 12,435member
    Quote:

    Originally posted by wmf

    Of course, you don't have to use a Mac to watch your (hypothetical) H.264 movies downloaded from the (hypothetical) iTunes Movie Store. You could use a (hypothetical) EyeHome Extreme based on the SMP8630 chip, which would be much cheaper than putting a Mac on top of your TV.



    True. I don't expect that Apple would use a IBM or Motorola CPU in a STB. I think we need a special chip that is designed to accelerate encode/decode of codecs. I believe you're referencing the Sigma chip right? I see a lot of activity from companies like Broadcom and Sigma in this area. It'll only be a matter of time. I think Cell is likely to be a wee bit expensive at first. We'll see but it's obvious that AVC at bettr than SD quality takes some crunching.
  • Reply 54 of 59
    123123 Posts: 278member
    Quote:

    Originally posted by danielctull

    I honestly can't believe you guys are questioning it!



    The QT one is more pixelated and has more compression artifacts, but it's also more detailed (as some of the higher frequencies are preserved), offers better contrast, and hasn't been put through an extreme low pass as the other one.



    But as I said, the whole thing is completely useless anyway as a) the source image isn't provided, and b) different players with different post processing are used to render the images.
  • Reply 55 of 59
    dfilerdfiler Posts: 3,420member
    Quote:

    Originally posted by 123

    But as I said, the whole thing is completely useless anyway as a) the source image isn't provided, and b) different players with different post processing are used to render the images.



    Useless? Perhaps from an academic standpoint. Some of us are also interested in what we'll see when actually watching a video in real life.



    I finally got done ripping my first DVD into h264. Handbrake worked flawlessly except for taking over 24 hours on my dual 1.4 (100mhz bus). It averaged around 10fps for quite a while and then 5fps for the rest of the rip. Sorry, I wasn't paying attention to which 'stage' was what and how long each took.



    The image, as noted earlier, turned out a bit dark for viewing on my computer's display. However, when piped through a television, the gama levels seemed to be spot on. There were no visible artifacts even when the DVD was compressed down to 700megs with 5.1 aac128 audio.



    Playback is quite processor intensive at 720x400, eating up around half of each CPU. Scrubing is also much slower, resulting in a two second lag when jumping to another spot in the film.



    One nice thing about QT7 (and the new dvd player) is that live resizing actually works. Although it isn't glass smooth when playing back h264 at this size on my computer, it is amazingly fast. I'm unsure what the iChat preview window is displayed in but it resizes absolutely flawlessly.



    In particular, i noticed less banding, blocking, and color artifacts as compared to a similar divx rip. It looks like divx and h264 have enough followers and active developement to result in some really great competition over the next few years.
  • Reply 56 of 59
    Quote:

    It looks like divx and h264 have enough followers and active developement to result in some really great competition over the next few years.



    Over at DivX Labs they are working on DivX Fusion which looks really interesting. It is able to hold multiple movies and dvd-like menus in one file. I don't know whether I would like to navigate through menus rather than just clicking the file to play it, but it does sound interesting and look interesting, although it does crash my PC.
  • Reply 57 of 59
    Quote:

    Originally posted by hmurchison

    True. I don't expect that Apple would use a IBM or Motorola CPU in a STB. I think we need a special chip that is designed to accelerate encode/decode of codecs. I believe you're referencing the Sigma chip right? I see a lot of activity from companies like Broadcom and Sigma in this area. It'll only be a matter of time. I think Cell is likely to be a wee bit expensive at first. We'll see but it's obvious that AVC at bettr than SD quality takes some crunching.



    from MOSR :
    Quote:

    Apple has signed an agreement with Alphamosaic, a recently acquired division of Broadcom to use Alphamosaic's forthcoming (end-of-year) VC02 (VideoCore 2) embedded multimedia processor -- a low-cost way to power devices that feature high-resolution, pixel-shaded 3D interfaces and H.264 video.



  • Reply 58 of 59
    brendonbrendon Posts: 642member
    I'm surprised that the word RAYCER has not crept up. This was a chip designer that Apple bought back several years ago. Maybe they designed a special chip for Apple that they will include on the new machines.
  • Reply 59 of 59
    danielctulldanielctull Posts: 586member
    Quote:

    Originally posted by ZO

    funny you should mention the save thing. I then tried to save the file with a track disabled and simply told me "unable to save". Gonna have to report that. Dammit.



    Still not sorted out on release... I hope they sort this out soon.



    All of my H.264 files playback much smoother on QT7, however I have had instances where old "normal" mov files are choppy, which they never were before. I'm still on 10.3.9 at the moment, hopefully a nice clean system with Tiger will speed it up some more.
Sign In or Register to comment.