VideoHelp Forum
+ Reply to Thread
Page 3 of 3
FirstFirst 1 2 3
Results 61 to 76 of 76
Thread
  1. Originally Posted by Epaminaidos View Post
    And even if, I would probably not have the patience for that. On my hardware, encoding a full 2-hour-movie in x.264 placebo would take more than 60 hours (given that "tune ssim" does not accelerate the encode).

    BTW: What is so important about "tune ssim"? Feels a bit like cheating to get a good rating.

    Does this really invalidate a comparison of codecs?

    I just opened a few movies and hand-counted the duration between the camera-switches (how do you call that duration?). The average was about 6-7s only. Very much of the previous shot cannot be used anyway I guess.
    Of course it invalidates a codec comparison, in broad terms here's what an encoder does:

    I analyses the video, both spatially (within a frame) and temporally (between frames) to find and decide what details to throw away and what details to keep. It uses all sorts of analysis techniques, including motion vectors and then it goes about encoding the video. It also decides where and how many I-Frames to place, if you take a 500 frame clip with the default x264 settings, you have already stated 2 I-Frames max (it uses a GOP length of 250, though I think at one point it was defaulting to 500), you have effectively taken scene-cut out of the equation (which decides where a key-frame, which is an I frame, should be placed) you have severely limited the B-frame decision making and all the other algorithms; this holds true for a single pass as well, including that silly encoding mode CRF, which is why the x264 developers never specify it for codec comparisons.

    Tune-SSIM IS a cheating mode, just like tune-PSNR, they even state as much in their description.

    And of course I don't expect you to encode a 2 hour movie in with x264 placebo, I would throw my computer out the window if I had to do that myself.

    But here's what you could do, take a standard DVD, one that's a high quality DVD, rip it to your hard drive, use AviDemux to cut out a 10 minute segment (choose a segment at random), then encode that with x264+medium, x265+medium and QSVenc, on a system as fast as yours that should take no time at all.

    I think most people would be happy if the Intel's encoder can beat either x264 or x265 with the medium presets.

    Also, when encoding with QSVenc, install and run GPU-Z to make sure that the video engine is actually being used, i.e. that it's not in software mode.
    Quote Quote  
  2. Member
    Join Date
    Sep 2016
    Location
    Deutschland
    Search PM
    Originally Posted by sophisticles View Post
    I analyses the video, both spatially (within a frame) and temporally (between frames) to find and decide what details to throw away and what details to keep. It uses all sorts of analysis techniques, including motion vectors and then it goes about encoding the video. It also decides where and how many I-Frames to place, if you take a 500 frame clip with the default x264 settings, you have already stated 2 I-Frames max (it uses a GOP length of 250, though I think at one point it was defaulting to 500), you have effectively taken scene-cut out of the equation (which decides where a key-frame, which is an I frame, should be placed) you have severely limited the B-frame decision making and all the other algorithms;
    Thank you for the explanation. I am not sure that I share your conclusion that this invalidates the whole comparison. I am by no means an expert in encoding-techniques. The last time, I checked it thoroughly, most cuts were used for keyframes anyway and not too much of the content before the cut could be used to encode the content after the cut. Thus I do not expect too many differences between a 10s-clip without cut and a whole movie with cuts every 5-10 seconds.
    But we could probably argue about that for ages without coming to a conclusion. Only empirical data might help a little, which we do not have.
    But here's what you could do, take a standard DVD, one that's a high quality DVD
    Believe it or not: I do not own a single DVD
    Quote Quote  
  3. Originally Posted by Epaminaidos View Post
    BTW: What is so important about "tune ssim"?
    In theory, x264 and x265 use techniques that optimize perceived visual quality but result in lower ssim scores. --tune ssim disables those particular optimizations.
    Quote Quote  
  4. Dinosaur Supervisor KarMa's Avatar
    Join Date
    Jul 2015
    Location
    US
    Search Comp PM
    Originally Posted by Epaminaidos View Post
    Originally Posted by sophisticles View Post
    But here's what you could do, take a standard DVD, one that's a high quality DVD
    Believe it or not: I do not own a single DVD
    As this thread seems to have stagnated over vague video source demands and movement of goal posts I decided to find a longer high quality source and share it.

    The below source comes off a 2009 Samsung Demo Bluray Disc designed to test your home theater system or to be played on the TVs for sale at a store. It's a short movie trailer encoded at 25Mbit in H.264. It also came with DTS-MA surrond sound which I removed by not including when muxing the original .M2TS into .MKV. The only problem with the source is that it has NTSC Telecine (to play Film at NTSC frame rates). Which can be corrected with the below avisynth script.

    The below link contains 3 files, archived and split with 7zip. I split the file to make it easier for me to upload. You just need to download all 3 files to the same directory and open the first file with 7zip, Winrar, Peazip, or whatever. As the copyright on this content might be a bit iffy, even though it's just an advertisement, I password protected the archive. The password is "videohelp" without the " ".
    https://jumpshare.com/b/wOmo4J5NFp8B5PjXlnVJ

    To remove the telecine in the source I used the below avisnyth script. Which can be used to make a lossless video via Virtualdub, if you can't get Avisynth to feed directly to QuickSync. Staxrip should be able to feed avisynth directly to QSVEncC.

    Code:
    LoadPlugin("O:\Megui\Megui\tools\lsmash\LSMASHSource.dll")
    LoadPlugin("O:\Megui\Megui\tools\avisynth_plugin\TIVTC.dll")
    LWLibavVideoSource("O:\........\Trailer Test Source.mkv")
    tfm(order=-1)
    tdecimate()
    I did some test encodes using different settings and codecs, besides QuickSync which hopefully Epaminaidos can help with. I aimed for 2000kbps in all my tests with CRF, by having to re-encode until coming close to 2000kbps. I'll post my test videos in my next post.
    Last edited by KarMa; 10th Oct 2016 at 02:02. Reason: Small error
    Quote Quote  
  5. Dinosaur Supervisor KarMa's Avatar
    Join Date
    Jul 2015
    Location
    US
    Search Comp PM
    To save Epaminaidos time, I encoded my test video myself and provide the test clips in the link below. And maybe Epaminaidos can provide a Quicksync encoding. I used Slower instead of Placebo as I had to encode the video multiple times to get the desired bitrate, usually 3-4 times. There really shouldn't be much of a difference between the two besides encoding time. All tests left the resolution the same as the source.

    x265 - Preset Slower - CRF 22.5 - ~2000kbps
    x265 - Preset Slower - CRF 28.7 - Tune Grain - ~2000kbps
    x264 - Preset Very Slow - CRF 25.1 - ~2000kbps
    Xvid - Megui HQ 2 Pass Preset- ~2000kbps

    Password is "videohelp" without the " ".
    https://jumpshare.com/b/X3EMU63mqzBxh8LRXQ0Q

    For fun I also tried encoding with HCEnc for a MPEG2 sample but I could barely get it below 3000 and the quality was rich in psychedelic artifacts.
    Quote Quote  
  6. Member
    Join Date
    Sep 2016
    Location
    Deutschland
    Search PM
    Sorry, I did not visit for a few days. Just found your reply.

    I encoded the files using the following command-line (QSVEnc 2.56)
    %qsvEnc% -i %original% -o "%targetDirectory%\%filename%" --codec hevc --vbr %vbr% --quality %preset%

    When setting the bitrate to 2000, QSVEnc sets the target AND max bitrate to 2000 and the output had only 1800. Thus I used a vbr with 2200kbps, which led to about 2000kbps.

    This created some warnings:
    B pyramid is not supported on current platform, disabled.
    cop.AUDelimiter value changed off -> auto by driver
    cop3.DirectBiasAdjustment value changed off -> auto by driver
    cop3.GlobalMotionBiasAdjustment value changed off -> auto by driver

    According to the spec, there are 6 quality-settings: best, higher, high, balanced, faster, fastest
    But when encoding, some files are exactly the same in a bitwise comparison:
    - fastest and faster
    - fast, balanced and high
    - higher and best

    There seem to be only three settings internally.

    Thus I only uploaded three files to http://jmp.sh/5Rpl4wm

    You know the password.

    Encoding was very fast:
    - fastest: 33s
    - balanced: 61s
    - best: 102s

    About the quality: A little bit better than XVID, but much worse than x264 and x265.
    Quote Quote  
  7. Dinosaur Supervisor KarMa's Avatar
    Join Date
    Jul 2015
    Location
    US
    Search Comp PM
    Thank you for the encodings but your videos still has the same NTSC Telecine as the source, which is why you can see combing artifacts in the scenes with movement. My avisynth script takes that out (post #64) by field matching and by removing duplicate frames, changing the frame rate from 29.97 to 23.976.

    I think I'll just upload a progressive source to make it easier on everyone. Either I might make a high quality re-encoding of my first source with the telecine removed (which hurts quality slightly), or just pick a naturally progressive source from somewhere else.

    Anyway thanks for the effort.
    Last edited by KarMa; 21st Oct 2016 at 21:05. Reason: grammer
    Quote Quote  
  8. Member vhelp's Avatar
    Join Date
    Mar 2001
    Location
    New York
    Search Comp PM
    I didn't particularly like the quality of those quicksync encodes. I think my encode of that trailer came out better. Not criticizing, just saying in my opinion. Anyway.

    Are there screenshots of the quicksync encode user-configuration UI console ?

    I'd be curious what users-configurations for encoding are there. Does it have crf and other rate controls to configure? How far can you go with tweaking, different things, etc.
    Quote Quote  
  9. Dinosaur Supervisor KarMa's Avatar
    Join Date
    Jul 2015
    Location
    US
    Search Comp PM
    Originally Posted by vhelp View Post
    I didn't particularly like the quality of those quicksync encodes. I think my encode of that trailer came out better. Not criticizing, just saying in my opinion. Anyway.
    He left the telecine in. So yeah, it's going to look bad and don't consider it a valid representation of quick sync. I'm getting a progressive source so hopefully Epaminaidos will be up for encoding another source.
    Quote Quote  
  10. Dinosaur Supervisor KarMa's Avatar
    Join Date
    Jul 2015
    Location
    US
    Search Comp PM
    Sorry for the delay but I got around to getting some progressive high quality sources, which allows for a simple encoding without need for deinterlacing or detelecining. As with my first source, these also come from demo discs designed to test out equipment or be played on the TV at the store. The audio has been stripped and remuxed into an MKV on both, to help it stay within the US fair use doctrine (research).

    Test Source 2: Real World footage, captured on 70mm film and transfered at 4K (to be downscaled to 1080p). I'm currently making 2,000kbps 1080p encodings for this one with x265.
    https://jumpshare.com/b/yzVF79diLAHcX8MPEkL0

    Test Source 3: Well known kids animation on film, with lots of stagnant backgrounds for easy compression. I'm currently making 1,000kbps 1080p encodings for this one with x265.
    https://jumpshare.com/b/ltIrjDuGFNLxJYm8jJNl

    After downloading one of the set of 3 files, open the first file with 7zip/Winrar/Peazip to extract it. Password is "videohelp" without the " ".
    Last edited by KarMa; 21st Oct 2016 at 21:10.
    Quote Quote  
  11. Member
    Join Date
    Sep 2016
    Location
    Deutschland
    Search PM
    Hi,

    I used the same settings as mentioned above to encode the test-sources.
    http://jmp.sh/5Rpl4wm

    The encoding-durations were about the same as well.
    Quote Quote  
  12. Dinosaur Supervisor KarMa's Avatar
    Join Date
    Jul 2015
    Location
    US
    Search Comp PM
    Originally Posted by Epaminaidos View Post
    Hi,

    I used the same settings as mentioned above to encode the test-sources.
    http://jmp.sh/5Rpl4wm

    The encoding-durations were about the same as well.
    Awesome stuff, I'll try to get some x265 encodings out soon.
    Quote Quote  
  13. Member
    Join Date
    Sep 2016
    Location
    Deutschland
    Search PM
    I should have read your post more thoroughly...
    I just uploaded Test 3 with a VBR-setting of 1.100 resulting in 1.000 kbps instead of 2.000
    Quote Quote  
  14. Dinosaur Supervisor KarMa's Avatar
    Join Date
    Jul 2015
    Location
    US
    Search Comp PM
    Originally Posted by Epaminaidos View Post
    I should have read your post more thoroughly...
    I just uploaded Test 3 with a VBR-setting of 1.100 resulting in 1.000 kbps instead of 2.000
    Even better thanks
    Quote Quote  
  15. Dinosaur Supervisor KarMa's Avatar
    Join Date
    Jul 2015
    Location
    US
    Search Comp PM
    Been thinking about how to go about comparing these. Tried doing screenshotcomparison.com but that was choking on all my images. So made my own script for jumping back and forth between the same frames in the two videos. Comparing every 200 frames. In the interest of keeping things clean, I used lossless Lagarith with the NULL enabled. For these specific comparisons the NULL feature saves a tremendous amount of space but VLC seems to have problems with the NULL frames. MPC-HC plays it well, and I get the best results in Virtualdub (but be sure to install the Lagarith codec if you have not).

    To compare the provided QuickSync HEVC to x265, I used x265 in single pass ABR mode, and tuned to Grain with the Slower preset. CRF or 2-Pass would probably give better results but figured for the first run I would do ABR. I also gave x265 200 frame look ahead, and allowing for up to 1000 frame GOP.

    The encodings took forever on my FX-6300. In grain tuning it was like half a frame a second, and without any tuning it was about one frame a second.
    Image Attached Files
    Quote Quote  
  16. Dinosaur Supervisor KarMa's Avatar
    Join Date
    Jul 2015
    Location
    US
    Search Comp PM
    My x265 encodings used to make the above comparison, password is videohelp. https://jumpshare.com/v/Tsbyqt868cWvnZtcjjBV?b=q4IlmQELB4razIPKH3dA


    Epaminaidos QS encodings can be found in post #71.
    Last edited by KarMa; 16th Nov 2016 at 22:03.
    Quote Quote  



Similar Threads

Visit our sponsor! Try DVDFab and backup Blu-rays!