VideoHelp Forum
+ Reply to Thread
Results 1 to 17 of 17
Thread
  1. ENTJ DrDeceit's Avatar
    Join Date
    Jul 2012
    Location
    United States
    Search PM
    For a movie, such as Avengers (2012), source is the untouched Blu-ray video
    Re-encoding it at 1920x1080 or 1280x720 resolutions

    I have a few questions, Subme 9 is RD of course, subme 10 is Quantisized Psychovisual Rate Distortion in all frames, 11 is "Full RD"

    Does subme 11 really do anything? I heard it was for a newer trellis, How does subme 9 10 11 differ in visual quality and compression? How do they work?

    Merange 16 vs merange 32 vs 64, etc. I found merange 32 isnt much slower than 16, but whats the compression and quality benefit, is it worth it?

    I encode 1080p or 720p versions of my bluray movies to fit on DVD5/DVD9 discs (BD5/BD9 Discs) using FULLY compatible options, I don't even enable weighted p-frames, to ensure compatibility with all players, EVEN THOUGH my player can play any AVC video, any resolution, any buffsize, any level, etc.

    I use 4 ref frames for 1080p, 6 for 720p (max)
    3 b frames

    CABAC
    8x8 transform
    no weighted p frames

    Pyramidal B-frames Strict

    Adaptive B-frames Optimal
    Adaptive direct mode Automatic

    Uneven Multi Hexagon
    subme 10 QPRD in all frames
    Motion estimation range 32

    Adaptive Quantization Strength 1.00
    Psy Rd 1.00
    Psychovisual Trellis 0.15

    Partitions All
    Trellis Always

    Deblock -3,-3 or -1,-1 or 0,0 (most often -1,-1)

    2 pass, turbo first pass
    Max bitrate 14.0 Mbps, bufsize 15.0 Mbps
    RC lookahead 0, no mbtree, keyint 48, keyint min 1, qpmin 10, qpmax 51

    cabac=1 / ref=6 / deblock=1:-1:-1 / analyse=0x3:0x133 / me=umh / subme=10 / psy=1 / psy_rd=1.00:0.15 / mixed_ref=1 / me_range=32 / chroma_me=1 / trellis=2 / 8x8dct=1 / cqm=0 / deadzone=21,11 / fast_pskip=1 / chroma_qp_offset=-3 / threads=3 / sliced_threads=0 / nr=0 / decimate=1 / interlaced=0 / bluray_compat=1 / constrained_intra=0 / bframes=3 / b_pyramid=1 / b_adapt=2 / b_bias=0 / direct=3 / weightb=1 / open_gop=1 / weightp=0 / keyint=48 / keyint_min=1 / scenecut=40 / intra_refresh=0 / rc_lookahead=0 / rc=2pass / mbtree=0 / bitrate=3400 / ratetol=1.0 / qcomp=0.50 / qpmin=10 / qpmax=51 / qpstep=4 / cplxblur=20.0 / qblur=0.5 / vbv_maxrate=14000 / vbv_bufsize=15000 / nal_hrd=vbr / ip_ratio=1.10 / pb_ratio=1.10 / aq=1:1.00
    Last edited by DrDeceit; 18th Nov 2012 at 12:01.
    Quote Quote  
  2. ENTJ DrDeceit's Avatar
    Join Date
    Jul 2012
    Location
    United States
    Search PM
    anyone?
    Quote Quote  
  3. Generally it's not worth it, except in some cases like very low bitrate encodes .

    Subme - Quantitatively , there is a tiny fractional benfit in SSIM for subme 11 vs. 10 vs. 9 when testing on a variety of sources . Qualitatively, sometimes I've seen cases where the quality is actually slightly worse

    Merange - anything over 24 is overkill. Even "placebo" has merange at 24.
    Quote Quote  
  4. I spent some time recently playing around with MSU's Video Quality Measurement Tool with x264 and have found that settings such as "me_range", "subme" and "me" have very little effect on the visual quality.

    The temptation of course is to set them to "full", but IMHO the loss of encoding speed compared to the negligable gain in quality is just not worth it.

    Below is what I got when I compared two files (720x576 DVD source) with the following settings:

    Code:
    --me esa --subme 10 --merange 24 --crf 20 --- Speed - 19.5 fps
    --me hex --subme 7 --merange 14 --crf 20 --- Speed - 57.5 fps
    Click image for larger version

Name:	LQ vs HQ.png
Views:	528
Size:	11.0 KB
ID:	14911

    Not much difference there, certainly nothing I could notice with the naked eye.

    If we compare the same high quality settings with medium settings with a small hike in --crf.

    Code:
    --me umh --subme 7 --merange 20 --crf 19.5 --- Speed - 48.0 fps
    --me esa --subme 10 --merange 24 --crf 20 --- Speed - 19.5 fps
    Click image for larger version

Name:	HQ vs CRF.png
Views:	460
Size:	11.2 KB
ID:	14912

    This seems to suggest that --crf has much more of an effect on quality than anything else.

    Of course the lower the --crf, the bigger the file size, but the 2.5x speed increase seems like a good trade-off.
    Last edited by mh2360; 29th Nov 2012 at 11:23.
    Quote Quote  
  5. mh2360 - can you add the resulting filesizes ? You can't compare "quality" at different bitrates . You either have to do multiple incremental CRF encodes or 2pass bitrate . So at a given bitrate, higher --merange and --subme will yield slightly better quality

    They do make minor differences, but the effect is larger and noticable on very low bitrate encodes . For "typical" scenarios, it's usually not worth it, and especially if you are using CRF (because usually you don't pay as much attention to the filesize or bitrate)

    A larger value for merange can make a bigger difference on HD encodes with lots of action compared to a SD source (but overall it's still very tiny difference)
    Quote Quote  
  6. Of course you are correct in saying that you can't compare quality at different bitrates, so I compared another file (The World War Z trailer downscaled to 720x576 if anyone is interested), at a much lower bitrate using 2 passes.

    The settings used for both samples are below.

    Code:
    --me hex --subme 6 --merange 14  --bitrate 1400 --trellis 0
    
    --me esa --subme 10 --merange 24  --bitrate 1400 --trellis 2
    Click image for larger version

Name:	1400kbps.png
Views:	449
Size:	9.8 KB
ID:	14913

    1400kbps is below what I would use for a DVD conversion, but there's still very little difference, both visually and according to the chart.

    It could be that I'm using SD sources, and that using higher settings might be more helpful with HD material at lower bitrates.
    Last edited by mh2360; 29th Nov 2012 at 13:49.
    Quote Quote  
  7. Even when we drop the bitrate to 1000kbps, we see that the average SSIM is dropping, but there's still very little difference between the samples. Visually I can see no difference at all.

    Click image for larger version

Name:	1000kbps.png
Views:	418
Size:	9.9 KB
ID:	14914

    Must do some HD testing...
    Quote Quote  
  8. Originally Posted by mh2360 View Post
    there's still very little difference, both visually and according to the chart.

    It could be that I'm using SD sources
    I don't think it makes any difference. Except for the fact that when you watch an SD source full screen the errors are magnified that much more, compared to watching an HD video of the same "quality".
    Quote Quote  
  9. There is definitely a visible quality difference in the lower bitrate range with the lower subme values vs. higher subme values, when viewing still frames.

    At "normal" or higher bitrates, or normal viewing - it won't be very noticible

    Difference between subme 7 vs. something like 9 or 10 is minimal . The OP was considering 9-11 and the difference between these is negligible (only visible through metrics, or the odd frame). But difference between something like 2-3 vs. 9-10 is definitely noticeable and reproducable on almost all sources in the low bitrate range. b-frames specifically suffer the effects, they are noticably worse . I've done many, many , many tests that show this. I can dig up some old examples - there should be some posted either here or doom9

    Also - Be careful about using SSIM or PSNR to measure "quality" . This has been discussed many times before
    Quote Quote  
  10. What I think goes through peoples' mind when they set up the x264 encoder is that they see how much slower the encoder runs when everything is set to "full", and naturally they assume that all that extra processing must mean that it's making a huge difference quality-wise.

    What I attempted to show is that just because x264 is running 2.5x slower with certain settings, the quality isn't going to be 2.5x better (or even 0.01x better). For most jobs the default settings are just as good as the HQ ones, if not better since the encoder will run much faster.
    Quote Quote  
  11. Originally Posted by mh2360 View Post
    What I think goes through peoples' mind when they set up the x264 encoder is that they see how much slower the encoder runs when everything is set to "full", and naturally they assume that all that extra processing must mean that it's making a huge difference quality-wise.

    What I attempted to show is that just because x264 is running 2.5x slower with certain settings, the quality isn't going to be 2.5x better (or even 0.01x better). For most jobs the default settings are just as good as the HQ ones, if not better since the encoder will run much faster.
    Yes, definitely. The settings used for "placebo" are aptly named

    It is a tradeoff and to most people and "normal" usage scenarios - the slower settings aren't worth it

    For pixel peepers, there is a noticable difference as you go into slower settings in low bitrate scenarios. "low bitrate" meaning relative to content complexity. It's almost never noticable in motion, but on frame-by frame examination it will be noticable. Higher bitrates alleviate everything and make differences difficult to visualize. The differences are very clear in the lower bitrate ranges, but almost imperceptible if you use low quantizers/low crf or high bitrates

    The OP was considering merange 64 . Well "placebo" has it set at 24.... so that tells you something. Only on a 4k action packed movie might you have a chance to catch a pixel that might have been missed at a lower value. It's exceedingly rare and the cost/benefit ratio is very very very low . For SD resolutions, don't even think about it
    Quote Quote  
  12. @mh2360:
    What tools do you use to produce out your charts? It would help me a lot with my coding testing. TIA
    Quote Quote  
  13. Originally Posted by saribex View Post
    @mh2360:
    What tools do you use to produce out your charts? It would help me a lot with my coding testing. TIA
    Those charts are from MSU VQMT
    Quote Quote  
  14. Originally Posted by mh2360 View Post
    I spent some time recently playing around with MSU's Video Quality Measurement Tool with x264 and have found that settings such as "me_range", "subme" and "me" have very little effect on the visual quality.
    There is one place where me and subme have a pretty obvious effect: dark grainy shots. Low me and subme values will lead to visible posterization. Here's are two examples, one encoded with the slow preset (slow.mkv), the other with the slow preset but me and subme turned down with --me=dia --subme=1 (slow-.mkv), both at CRF=18.
    Image Attached Files
    Quote Quote  
  15. @jagabo - you're replying to a mh2360's 2 year old post

    Here's another example, 720p 8Mbps, subme 0 + dia vs. subme 9 + multihex . Differences are even more apparent at lower bitrates
    https://forum.videohelp.com/threads/353166-Fast-encoding-with-x264
    Quote Quote  
  16. Originally Posted by poisondeathray View Post
    @jagabo - you're replying to a mh2360's 2 year old post
    I know. I just thought an example would be useful since there were none in this thread. And that example was at a CRF where people often consider x264 to be nearly "transparent" at normal playback speed.
    Quote Quote  
  17. Okay I've installed the MSQ Video Quality Measurement Tool + newest version of avisyth .. but can't figure out how to compare my raw .m2ts file from a blu-ray with my .mkv re-code with x264... Can anybody help me step by step? MSU Tools seems not to recognize any of both files...
    TIA guys
    Quote Quote  



Similar Threads

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