VideoHelp Forum
+ Reply to Thread
Page 24 of 27
FirstFirst ... 14 22 23 24 25 26 ... LastLast
Results 691 to 720 of 782
Thread
  1. Originally Posted by aymanalz View Post
    Originally Posted by -Habanero- View Post
    Originally Posted by sneaker View Post
    Oh, I did the same test last year.
    And as predicted, your encode is better quality than mine despite being way older. Sigh.
    WHAT. AM. I. DOING. WRONG?
    Well first of all, a CRF of 32. That will look terrible. Lower it. Or use 2-pass encode at a certain bitrate for both x264 and x265, and then compare quality. Chances are, x265 will look much better.

    Since you want to know how the two codecs compare at similar bitrates, that would be the sensible thing to do. CRF is not the same for both.
    The file sizes are the same for both. I did 2pass as well but it was even worse quality as I just showcased at the top. It's getting frustrating how you people are not even looking at the videos I post before lecturing me. It would take 3 seconds for you to download and notice they are both 1.7MB.

    sneakers, what is the exact procedure you used? I wanna try exactly whatever you did.
    Quote Quote  
  2. Member
    Join Date
    Aug 2013
    Location
    Central Germany
    Search PM
    @ gregoor:

    "GPU acceleration" and "quality HEVC encoding" usually don't match because HEVC encoding is way too complex for the rather restricted architecture of GPGPUs. A Skylake CPU with several cores using x265 will beat GPGPUs in quality/speed ratio and efficiency easily.

    Apart from that: This is a thread about x265, a pure software encoder. Your question about GPU encoders doesn't fit here anyway. It would have been better to create a separate thread for it...
    Quote Quote  
  3. Originally Posted by -Habanero- View Post
    sneakers, what is the exact procedure you used? I wanna try exactly whatever you did.
    I don't remember, unfortunately. You'll have to read the settings from MediaInfo. Knowing me, I probably just did --preset veryslow or placebo, nothing complicated.
    Quote Quote  
  4. Originally Posted by sneaker View Post
    Originally Posted by -Habanero- View Post
    sneakers, what is the exact procedure you used? I wanna try exactly whatever you did.
    I don't remember, unfortunately. You'll have to read the settings from MediaInfo. Knowing me, I probably just did --preset veryslow or placebo, nothing complicated.
    That's what I do too, preset veryslow and add 16 refs for best quality. Can you do the test again with x265 2.0 and post the video here? Also, what is your source of the cartoon? It's tinted differently than mine.
    Quote Quote  
  5. My source is the DVD so no color conversions were involved. With 2.0+10 preset veryslow:
    Image Attached Files
    Quote Quote  
  6. lol, now I can't tell the quality difference. Yours has some frames worse than mine and some better. Can you do me a big favor and cut out that specific clip from your DVD directstream and send me so I can do a proper comparison? My source is tinted and cropped.

    EDIT: Thanks. I'm getting more or less identical results now so it appears nothing is wrong with my x264 build. But I did do the same commandline on both Windows and Linux for a test and it produced different results. The quality is indistinguishable but it puzzles me why the frames would not be identical. According to SSIM the Linux encode is slightly better.

    Now time to do a process of elimination because if I'm using 16 refs on animation and getting worse quality than the default 5 refs then it's something else screwing it all up. So far I just use a higher lookahead and enable non-conformance. I mean with x264 I'm unable to get the file below 2,240 KB on 10-bit so I have a hunch it's the large lookahead I'm using. But if this lowers the quality then that's news to me. o.o
    Last edited by -Habanero-; 11th Aug 2016 at 03:10.
    Quote Quote  
  7. @ sneaker: oh pardon, thanks for giving me those infos nonetheless
    Quote Quote  
  8. Originally Posted by -Habanero- View Post
    I'm getting more or less identical results now so it appears nothing is wrong with my x264 build. But I did do the same commandline on both Windows and Linux for a test and it produced different results. The quality is indistinguishable but it puzzles me why the frames would not be identical. According to SSIM the Linux encode is slightly better.
    Was this native linux, or VM ? A possible explanation is that with a VM, the cores and thus default threads setting will usually be lower than using a native host. The lower the threads the higher the quality, detectable by metrics like SSIM . You can manually set threads to rule that out
    Quote Quote  
  9. Originally Posted by LigH.de View Post
    @ gregoor:
    "GPU acceleration" and "quality HEVC encoding" usually don't match because HEVC encoding is way too complex for the rather restricted architecture of GPGPUs. A Skylake CPU with several cores using x265 will beat GPGPUs in quality/speed ratio and efficiency easily.
    That's really funny, HEVC was envisioned from the get go to be gpu accelerated. It's not that "HEVC encoding is way too complex for the rather restricted architecture of GPGPUs", this is a silly assertion considering gpu's are used to accelerate all sorts of incredibly complex operations, such as fluid dynamics simulations, weather predictions, medical imaging and video filtering, the real reason for the lack of gpu acceleration is that companies have turned to ASIC's and FPGA's to accelerate encoding, because they offer much faster performance and very low power consumption.
    Quote Quote  
  10. Member
    Join Date
    Aug 2013
    Location
    Central Germany
    Search PM
    GPGPUs are great to accelerate highly parallelizable algorithms (means, there are many small routines which can be calculated rather independently of each other, like particle physics).

    But HEVC is not as highly parallelizable as you may wish. There are too many dependencies between parts of the whole algorithm, need to wait for intermediate results. You will certainly have read already about a rather low CPU utilization for systems with a lot of logical cores. If you even have a dual socket system, the most efficient strategy may be running several instances of the encoder in parallel.

    And on top, there is always the overhead to move the content of video frames from the system RAM to the VRAM before the GPGPU units can access them, which may only be void if the graphic chipset already decodes the source video using an integrated decoder chip, but then you won't have an advantage if you prefer to encode already decoded video (which may be pre-filteres by e.g. AviSynth scripts).

    If you don't trust my "halfwit", then please wait for affirmative replies from the x265 developers. Of course they will know a lot better than me why they decided to develop a CPU-only encoder. Anything else will not be "x265", according to the current schedule, and should rather be discussed in a separate thread.
    Quote Quote  
  11. Originally Posted by LigH.de View Post
    If you don't trust my "halfwit", then please wait for affirmative replies from the x265 developers. Of course they will know a lot better than me why they decided to develop a CPU-only encoder. Anything else will not be "x265", according to the current schedule, and should rather be discussed in a separate thread.
    The x265 spokesperson, Tom, has said numerous times that they planned to bring gpu acceleration to x265 and bragged how they were behind OCL acceleration for x264. Tom has since changed his claims a bit, but I think the reason all codec creators, including all the HEVC codecs and VP9, focus on software only implementations is so that it's usable by a wide a user base as possible, GPGPU seems to require specialized code for each gpu architecture which means that even if they wrote major portions of it for a Pascal gpu it probably wouldn't work as well for Maxwell gpu's and not at all for AMD gpu's
    Quote Quote  
  12. PDR, you might be right, I didn't think of that. It is a VM and I do have it set to the same number of cores as my host so not sure. I did a bunch of tests and I'm thoroughly confused.

    0.96667 TouhouSource2 x264CRF.avs (my CRF x264 encode)
    0.96401 Touhoux265sneaker.avs (Sneaker's x265 encode)
    0.96391 TouhouSource2noTuning.avs (x265 encode using just --preset veryslow)
    0.96366 TouhouSource2 16ref.avs (x265 using above with 16 refs)
    0.96342 TouhouSource2 16bframes.avs (x265 with 16bframes with 5 refs)
    0.96332 TouhouSource2 16ref 100rclookahead nonconf.avs (exactly what it says)
    0.96306 TouhouSource2 16ref nonconf.avs (default rclookahead)
    0.96296 TouhouSource2 x264 2pass.avs (my 2pass x264 encode)
    0.95575 Touhoux264sneaker.avs (Sneaker's 2pass x264 encode)

    So using 16 refs vs. 5 produces LOWER quality? What the actual ****? This encoder is seriously messed up. The less I messed with the script, the better quality I got. The video does seem to benefit with a higher rc-lookahead tho. No idea why less b-frames is higher quality and I sure as **** have no idea why my x264 encode is still visibly higher quality than the best of these scenarios or why it's way higher quality than Sneaker's despite the only difference being him using psychovisuals while I don't.

    Looking forward to hear explanation from x265 development on this. o_O
    Quote Quote  
  13. Member
    Join Date
    Aug 2013
    Location
    Central Germany
    Search PM
    Do you believe that a higher SSIM value means better subjective quality? Think again, this time remembering psycho-visual enhancements...
    Quote Quote  
  14. I have surveyed the quality with my own eyes and the x264 encode is visibly better than the best x265 one. This wasn't the case last year. In every single test so far x265 has been doing worse than its predecessor.

    EDIT: Here are the SSIMPlus scores. Consistent with the regular SSIM from what I see. x264 still owns all for this clip and every clip I tested with recent builds.

    91.2 TouhouSource2 x264CRF
    90.23 TouhouSource2 x264 2pass
    90 Touhoux265sneaker
    89.97 TouhouSource2x265noTuning 2pass
    89.95 TouhouSource2x265 2pass Windows
    89.87 TouhouSource2x265 notuning CRF
    89.85 TouhouSource2 16ref
    89.82 TouhouSource2 bframes16
    89.75 TouhouSource2 16ref 100rclook nonconf
    89.65 TouhouSource2 16ref nonconf
    Last edited by -Habanero-; 16th Aug 2016 at 01:51.
    Quote Quote  
  15. So I tested every milestone x265 build from version 1.0 to 2.0 to see where the hell things started going wrong and the results astonish me. x265 1.0 produced the best CRF quality and x265 1.2 produced the best 2pass quality (only ABR was available for 1.0 and 1.1).

    Basically, the older the x265 build, the better the quality. This is retarded.

    I've had some trouble getting the filesize to stay the same so it's a little hard to compare each individual build to each other but the trend I see is that since 1.0 the quality is on a slight downturn and starting with 1.8 the quality noticeably drops.

    Method x265 version Filesize SSIM+
    CRF 1 1781 94.27
    CRF 1.1 1779 94.08
    CRF 1.5 1773 93.89
    CRF 1.6 1768 93.88
    CRF 1.7 1771 93.84
    CRF 1 1730 93.8
    CRF 1.2 1763 93.71
    CRF 1.3 1759 93.7
    CRF 1.4 1756 93.67
    CRF 1.5 1743 93.62
    CRF 1.8 1767 93.5
    CRF 1.9 1761 92.85
    CRF 2 1769 92.48

    Method x265 version Filesize SSIM+
    2pass 1.5 1768 93.78
    2pass 1.2 1754 93.74
    2pass 1.6 1755 93.7
    2pass 1.3 1745 93.65
    2pass 1.4 1742 93.65
    2pass 1.7 1755 93.63
    2pass 1.5 1741 93.61
    2pass 1.8 1756 93.25
    2pass 1.9 1758 92.71
    2pass 2 1766 92.41
    2pass 2 1742 92.27
    ABR 1 1769 91.51
    ABR 1 1750 91.42
    ABR 1.1 1751 90.26

    Command lines attached.

    What are the devs doing that's driving x265 down the gutter?
    Image Attached Files
    Quote Quote  
  16. Not necessarily comparable using the same "preset"; because some things were changed/added in the presets over time such as --limit-refs , psy settings changed over time, weightb is different between 1.0 and 2.0 . Look at the FPS , about 1.8x faster from v1.0 to v2.0

    Thanks for posting the test results, nice to look at the progression
    Quote Quote  
  17. focus for the defaults settings probably changed from quality to speed to accommodate folks not using x265 since it seemed to slow,...
    users currently on my ignore list: deadrats, Stears555
    Quote Quote  
  18. Member
    Join Date
    Aug 2013
    Location
    Central Germany
    Search PM
    And after all, it is still possible that the default optimization for psycho-visual quality may produce worse SSIM metric results, yet look subjectively better somehow. To benchmark with optimal SSIM metric results, you would have to compare all encodes with "--tune ssim".
    Quote Quote  
  19. I thought the preset was fairly conservative in the settings. I'll redo the whole test if any of you can recommend a default commandline for me to use for all of them. I'm out of work again so I have nothing better to do.

    PDR, I disagree. The speed was 7 fps on 1.2 for the second pass and better quality than 4.5 fps 1.9.

    Selur, that's what I'm thinking too. But if that's true, then why use x265 at all given x264 still produces better quality than 2.0 and is WAY faster? With every test I've done with 2.0 x265 had worse quality than x264, both for cartoons and real footage.

    Also, I'm using SSIM+, not SSIM. It is far better tuned to correlate to perceptual quality. While I'm yet to finish the psy tests on another thread, the author of SSIM+ told me that psychovisuals were taken into account when developing the new metric.
    Quote Quote  
  20. Originally Posted by -Habanero- View Post
    PDR, I disagree. The speed was 7 fps on 1.2 for the second pass and better quality than 4.5 fps 1.9.
    Yes, I missed that. I was looking at v1.0 and v2.0

    then why use x265 at all given x264 still produces better quality than 2.0 and is WAY faster? With every test I've done with 2.0 x265 had worse quality than x264, both for cartoons and real footage.
    I haven't done that many tests with v2.x yet, but x265 still rocks at UHD resolutions. I would think it's still better at smooth / simple animation content too even for lower resolutions. Those are 2 areas where it was clearly better than x264

    Also, you need a CPU with AVX2 for x265 to make use off most of the speed improvements (Haswell & newer)
    Quote Quote  
  21. Wow, I realized I forgot to include the 2.0 commandline for 2pass. It was even slower (3.69 fps) and the quality worse than 1.9. :/

    And the sample I'm using is the Touhou cartoon so it's ridiculous that x265 is doing so bad.

    If you guys don't recommend a commandline I'm just gonna do my own. So far you've recommended to not use psychovisuals. I'm down with that but before I use --tune ssim, what does this actually do? If the only thing it does is disable psychovisuals that's fine. But I don't want any artificial tinkering with the metrics. So what does it mess with besides psychovisuals?

    I also realized that builds went back and forth between using a fast and slow first pass which is why I had trouble controlling the filesize in 2pass mode. This would've been useful to know the first time.
    Quote Quote  
  22. I'm not sure how you would compare it properly, there are too many differences between versions

    --tune ssim for x265 is the same as for x264, it means --no-psy --aq-mode 2 . But in a sense that is "tinkering" with metrics because it's optimizing for them. Not very useful for extrapolating results for actual usage scenarios
    Quote Quote  
  23. The whole point is to compare the best of the capabilities of each codec which means taking advantage of whatever new setting each new version has. If the best of 2.0 outputs worse quality than the best of 1.0 despite all these new features, that's further proof that there's problems with the code.

    But because of the computational complexity, I can't use maximum settings so the question is what the proper 'medium' settings are. No matter, I will turn psychovisuals off.
    Quote Quote  
  24. I haven't done enough tests yet, but my general impression is that quality is definitely better than before than maybe a year ago. Speed is also definitely better

    At least you're trying to quantify these things; if I had more time I would do more thorough tests
    Quote Quote  
  25. For 480p content the quality doesn't look to be getting better.

    Anyway, I did a retest and this time the bitrates slightly increased linearly with each successive version yet the SSIM+ score was frozen despite new features.

    Method x265 version Filesize SSIM+
    2pass 1.2 1744 93.6
    2pass 1.3 1745 93.55
    2pass 1.4 1742 93.55
    2pass 1.5 1743 93.53
    2pass 1.6 1759 93.61
    2pass 1.7 1758 93.57
    2pass 1.8 1760 93.56
    2pass 1.8 1734 93.45
    2pass 1.9 1757 93.55
    2pass 2 1754 93.55

    See that? x265 1.2, 1744 KB got 93.6%, 1.6 which is 15 KB larger got the same score and 2.0 which is 10 KB larger got a slightly lower score despite the following new settings:

    1.3+ --psy-rdoq 0
    1.3+ --no-fast-intra
    1.4+ --deblock=1:1
    1.6+ --rdoq-level 2
    1.6+ --min-cu-size 8
    1.6+ --lookahead-slices 1
    1.8+ --limit-refs 1
    1.9+ --limit-modes
    1.9+ --rd-refine
    2.0+ --no-rskip

    So the oldest version capable of 2pass remains the quality winner. It's encoding speed was 6.25 fps and 2.0's encoding speed was... 3.35 fps. Twice as slower and worse quality. This is not looking good.

    Command lines:
    Code:
    avs4x26x --x26x-binary x265_12 "TouhouAnime.avs" --pass 1 --bitrate 67 --tune ssim --ref 1 --rd 2 --ctu 64 --no-rect --no-amp --early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 1 --me dia --subme 2 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao -o "Touhou Anime1.2 2pass.hevc"
    
    avs4x26x --x26x-binary x265_12 "TouhouAnime.avs" --pass 2 --bitrate 67 --tune ssim --ref 5 --rd 6 --ctu 64 --rect --amp --no-early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 4 --me star --subme 4 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao -o "Touhou Anime1.2 2pass.hevc"
    6.25 fps
    
    avs4x26x --x26x-binary x265_13 "TouhouAnime.avs" --pass 1 --bitrate 67 --tune ssim --ref 1 --rd 2 --ctu 64 --no-rect --no-amp --early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 1 --me dia --subme 2 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.3 2pass.hevc"
    
    avs4x26x --x26x-binary x265_13 "TouhouAnime.avs" --pass 2 --bitrate 66 --tune ssim --ref 5 --rd 6 --ctu 64 --rect --amp --no-early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 4 --me star --subme 4 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.3 2pass.hevc"
    5.89 fps
    
    avs4x26x --x26x-binary x265_14 "TouhouAnime.avs" --pass 1 --bitrate 67 --tune ssim --ref 1 --rd 2 --ctu 64 --no-rect --no-amp --early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 1 --me dia --subme 2 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.4 2pass.hevc"
    
    avs4x26x --x26x-binary x265_14 "TouhouAnime.avs" --pass 2 --bitrate 66 --tune ssim --ref 5 --rd 6 --ctu 64 --rect --amp --no-early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 4 --me star --subme 4 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.4 2pass.hevc"
    5.45 fps
    
    avs4x26x --x26x-binary x265_15 "TouhouAnime.avs" --pass 1 --bitrate 67 --tune ssim --ref 1 --rd 2 --ctu 64 --no-rect --no-amp --early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 1 --me dia --subme 2 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.5 2pass.hevc"
    
    avs4x26x --x26x-binary x265_15 "TouhouAnime.avs" --pass 2 --bitrate 67 --tune ssim --ref 5 --rd 6 --ctu 64 --rect --amp --no-early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 4 --me star --subme 4 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.5 2pass.hevc"
    5.04 fps
    
    avs4x26x --x26x-binary x265_16 "TouhouAnime.avs" --pass 1 --bitrate 67 --tune ssim --ref 1 --rd 2 --ctu 64 --no-rect --no-amp --early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 1 --me dia --subme 2 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --min-cu-size 8 --rdoq-level 2 --lookahead-slices 1 --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.6 2pass.hevc"
    
    avs4x26x --x26x-binary x265_16 "TouhouAnime.avs" --pass 2 --bitrate 67 --tune ssim --ref 5 --rd 6 --ctu 64 --rect --amp --no-early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 4 --me star --subme 4 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --min-cu-size 8 --rdoq-level 2 --lookahead-slices 1 --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.6 2pass.hevc"
    5.14 fps
    
    avs4x26x --x26x-binary x265_17 "TouhouAnime.avs" --pass 1 --bitrate 67 --tune ssim --ref 1 --rd 2 --ctu 64 --no-rect --no-amp --early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 1 --me dia --subme 2 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --min-cu-size 8 --rdoq-level 2 --lookahead-slices 1 --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.7 2pass.hevc"
    
    avs4x26x --x26x-binary x265_17 "TouhouAnime.avs" --pass 2 --bitrate 67 --tune ssim --ref 5 --rd 6 --ctu 64 --rect --amp --no-early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 4 --me star --subme 4 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --min-cu-size 8 --rdoq-level 2 --lookahead-slices 1 --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.7 2pass.hevc"
    5.57 fps
    
    avs4x26x --x26x-binary x265_18 "TouhouAnime.avs" --pass 1 --bitrate 67 --tune ssim --ref 1 --rd 2 --ctu 64 --no-rect --no-amp --early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 1 --me dia --subme 2 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --limit-refs 1 --min-cu-size 8 --rdoq-level 2 --lookahead-slices 1 --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.8 2pass.hevc"
    
    avs4x26x --x26x-binary x265_18 "TouhouAnime.avs" --pass 2 --bitrate 66 --tune ssim --ref 5 --rd 6 --ctu 64 --rect --amp --no-early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 4 --me star --subme 4 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --min-cu-size 8 --rdoq-level 2 --lookahead-slices 1 --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.8 2pass 66.hevc"
    (finally it tells me ALL the unrecognized options, not just one at a time)
    6.64 fps
    
    avs4x26x --x26x-binary x265_19 "TouhouAnime.avs" --pass 1 --bitrate 67 --tune ssim --ref 1 --rd 2 --ctu 64 --no-rect --no-amp --early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 1 --me dia --subme 2 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --limit-modes --rd-refine --limit-refs 1 --min-cu-size 8 --rdoq-level 2 --lookahead-slices 1 --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.9 2pass.hevc"
    
    avs4x26x --x26x-binary x265_19 "TouhouAnime.avs" --pass 2 --bitrate 67 --tune ssim --ref 5 --rd 6 --ctu 64 --rect --amp --no-early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 4 --me star --subme 4 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --limit-modes --rd-refine --limit-refs 1 --min-cu-size 8 --rdoq-level 2 --lookahead-slices 1 --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime1.9 2pass.hevc"
    8.26 fps
    
    avs4x26x --x26x-binary x265 "TouhouAnime.avs" --pass 1 --bitrate 67 --tune ssim --ref 1 --rd 2 --ctu 64 --no-rect --no-amp --early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 1 --me dia --subme 2 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --no-rskip --limit-modes --rd-refine --limit-refs 1 --min-cu-size 8 --rdoq-level 2 --lookahead-slices 1 --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime2.0 2pass.hevc"
    
    avs4x26x --x26x-binary x265 "TouhouAnime.avs" --pass 2 --bitrate 67 --tune ssim --ref 5 --rd 6 --ctu 64 --rect --amp --no-early-skip --b-intra --no-tskip-fast --tu-intra-depth 3 --tu-inter-depth 3 --tskip --max-merge 4 --me star --subme 4 --merange 57 --weightp --weightb --rc-lookahead 40 --b-adapt 2 --bframes 8 --cutree --sao --no-rskip --limit-modes --rd-refine --limit-refs 1 --min-cu-size 8 --rdoq-level 2 --lookahead-slices 1 --deblock=1:1 --no-fast-intra --psy-rdoq 0 -o "Touhou Anime2.0 2pass.hevc"
    3.35 fps
    Quote Quote  
  26. Speed difference, ok - not sure why that is, but that needs looking at.

    But quality wise - Wouldn't the difference be too small to be statistically significant? 67kb/s isn't a lot of bitrate, and 0.05% delta isn't a lot ; especially with a metric that doesn't necessarily have high correlation with "quality", despite claims that it's better than it's predecessor SSIM
    Quote Quote  
  27. Defaults change so crf will do different things with the same command line for different versions,... (CRF is only quality indicator if all other settings and the source stays the same!)
    And if you create two 2pass files with the same output file size one with preset placebo one with preset ultrafast, do you expect the same output quality?
    users currently on my ignore list: deadrats, Stears555
    Quote Quote  
  28. PDR, 15 KB of difference is a notable difference for a bitrate so small. And SSIM+ was subjectively tested by several subjects who gave scores out of 10. It correlates very well. I posted the graphs in that VFR thread. SSIM+ uses the same scoring format to eliminate this confusion altogether. You are right that it probably can't be relied on to evaluate small differences, tho. So I'll post comparison pics in a sec.
    Also, keep in mind SSIM+ takes motion into account unlike the previous. It doesn't treat video as a bunch of individual pictures.

    Selur,
    My new test doesn't involve CRF at all and I used the same quick first pass for all of them so what are you complaining about? And I did not use presets at all, nor do the settings resemble placebo. Notice I'm only using 5 refs and subme 4?

    EDIT: Here it is. The difference is small but notice that 2.0 completely smears one of the lines in the brunette's hair and on the third frame there is more ringing artifacts on the trees.
    These frames were chosen by an RNG, btw. So I'm not cherry-picking.
    Image Attached Files
    Last edited by -Habanero-; 29th Aug 2016 at 13:28.
    Quote Quote  
  29. 67kb/s isn't a lot for 720x480 even if it's anime . They both look bad, but I agree 2.0 is probably worse overall if you judge by those still frames alone. I would hesitate to draw conclusions from that test, even if you want to only apply them to low bitrate scenarios with anime sources
    Quote Quote  
  30. I encourage others to do their own, of course. The reason I use low bitrates in testing is to make differences more obvious. If high bitrates are used, both would look great and be indistinguishable.

    I've done more tests now that we've established which version ranks the highest against the current.

    Very strong settings for 1.2 and 2.0 on both 2pass and CRF.

    Method x265 version Filesize SSIM+
    CRF 1.2 1776 94.1
    2pass 1.2 1763 93.94
    CRF 2 1776 93.76
    2pass 2 1754 93.66
    2pass x264 1763 92.76
    CRF x264 1774 92.72

    Both of them beat 2.0 hands down. If you think 0.5% is not a big difference, remember the VFR tests we did on that South Park episode where the VFR got only 0.1% higher than the 98.9% CFR and it took 9% more bitrate to get it up to the same score.
    The difference is negligible but noticeable.

    Finally x264 behaves as expected and got lower quality. Previously, all were tested at 10-bit while now they're 8-bit so I guess 10-bit x264 performs more strongly enough to beat x265. I don't have the patience to do 10-bit tests now. They take forever.

    Commandlines:
    Code:
    avs4x26x.exe --x26x-binary x265 TouhouAnime.avs --pass 1 --bitrate 67 --tune ssim --preset veryslow --ref 16 --bframes 16 --rc-lookahead 100 --allow-non-conformance --psy-rdoq 0 -o "TouhouAnime x265 2.0 max 2pass.hevc"
    
    avs4x26x.exe --x26x-binary x265 TouhouAnime.avs --pass 2 --bitrate 67 --tune ssim --preset veryslow --ref 16 --bframes 16 --rc-lookahead 100 --allow-non-conformance --psy-rdoq 0 -o "TouhouAnime x265 2.0 max 2pass.hevc"
    
    avs4x26x.exe --x26x-binary x265_12 TouhouAnime.avs --pass 1 --bitrate 67 --tune ssim --preset veryslow --ref 16 --bframes 16 --rc-lookahead 100 -o "TouhouAnime x265 1.2 max 2pass.hevc"
    
    avs4x26x.exe --x26x-binary x265_12 TouhouAnime.avs --pass 2 --bitrate 67 --tune ssim --preset veryslow --ref 16 --bframes 16 --rc-lookahead 100 -o "TouhouAnime x265 1.2 max 2pass.hevc"
    
    avs4x26x.exe --x26x-binary x265 TouhouAnime.avs --crf 32.1 --tune ssim --preset veryslow --ref 16 --bframes 16 --rc-lookahead 100 --allow-non-conformance --psy-rdoq 0 -o "TouhouAnime x265 2.0 max CRF.hevc"
    
    avs4x26x.exe --x26x-binary x265_12 TouhouAnime.avs --crf 32.1 --tune ssim --preset veryslow --ref 16 --bframes 16 --rc-lookahead 100 -o "TouhouAnime x265 1.2 max CRF.hevc"
    Quote Quote  



Similar Threads

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