VideoHelp Forum
+ Reply to Thread
Results 1 to 4 of 4
Thread
  1. Some people were discussing x264 CRF. The fact is you cannot compare same CRF at different settings because they yield different quality. CRF does not mean same quality; it' s just a method of rate control

    Here is one of the older examples - It's the Planet Earth trailer. I couldn't find the older encodes on my archived HDD's, so I just redid them with the current x264. Yadif(order=1, mode=0) to single rate deinterlace, spline36resize to 720x400 in this example. Ok you can also argue that it's a "trailer" and not "typical" content either... But do enough of these tests, and I assure you will see differences 99 times out of 100

    Typically, you will see lower b-frame quality, and problems where there is motion, expecially with fine details like grass, grain etc....(open the pics in different tabs and flip back & forth) . Using one of the quicker presets , those types details won't be preserved as well. These screenshots below are b-frame comparisons. I'm uploading the source clip and encodes and will update once they are finished

    110 original
    Click image for larger version

Name:	110 original.png
Views:	644
Size:	523.1 KB
ID:	31559

    110 superfast
    Click image for larger version

Name:	110 superfast.png
Views:	559
Size:	491.7 KB
ID:	31560

    110 slower
    Click image for larger version

Name:	110 slower.png
Views:	572
Size:	518.5 KB
ID:	31561

    738 original
    Click image for larger version

Name:	738 original.png
Views:	521
Size:	439.5 KB
ID:	31562

    738 superfast
    Click image for larger version

Name:	738 superfast.png
Views:	590
Size:	398.6 KB
ID:	31563

    738 slower
    Click image for larger version

Name:	738 slower.png
Views:	519
Size:	419.8 KB
ID:	31564

    1166 original
    Click image for larger version

Name:	1166 original.png
Views:	637
Size:	302.8 KB
ID:	31565

    1166 superfast
    Click image for larger version

Name:	1166 superfast.png
Views:	572
Size:	269.5 KB
ID:	31566

    1166 slower
    Click image for larger version

Name:	1166 slower.png
Views:	625
Size:	276.3 KB
ID:	31567

    source clip
    https://www.mediafire.com/?2wbaka7n57qb4x6
    Image Attached Files
    Quote Quote  
  2. Another example from a consumer camcorder a few years ago. I couldn't find the tests, so I redid them with current x264 as above. These aren't new findings - this has been established years ago with dozens of tests posted (and thousands probably not posted). I just can't seem to find the threads (maybe they were at doom10 which is down now of course...)

    The MTS original clip can be downloaded from vimeo if you are signed in
    Code:
    http://vimeo.com/10663338
    For the test it was resized to 1280x720p59.94 with spline36resize(), and just a 10 sec section was tested but feel free to repeat the test on a longer section

    Again - b-frame comparions: Notice the fine fur details are obliterated in the superfast encode, also that the file size "trend" is switched (slower encode is actually larger). If you look on other frames, the backround details are reduced as well, the moss is over smoothed for example. This detail loss is typical 99 times out of 100 when looking at quicker settings vs. slower settings for the vast majority of source scenarios. It's not a "cherry picked" example. The only time you might not see differences if source had few details to begin with, or there is very little motion, e.g. maybe some low quality animation


    original
    Click image for larger version

Name:	298 original.png
Views:	324
Size:	1.91 MB
ID:	31581

    superfast
    Click image for larger version

Name:	298 superfast.png
Views:	327
Size:	1.80 MB
ID:	31582

    slower
    Click image for larger version

Name:	298 slower.png
Views:	330
Size:	1.86 MB
ID:	31583
    Image Attached Files
    Quote Quote  
  3. Interesting, you were right the whole time. The one with weak settings does look worse despite the higher bitrate. But what did I do in my limited tests that was so different? The only major difference I see is that I had mb-tree enabled with both which might play a part. But I realize complaining about this invalidates what I originally thought anyway because a CRF was supposed to maintain the quality regardless of the settings.

    I also expected the ultraslow one to be lower quality. I vaguely remember around 2009 how I had to lower the CRF if I used very strong settings because x264 would under-allocate bits, the video would be twice as smaller than medium preset and looked slightly worse.
    Quote Quote  
  4. In order to see differences, you need more complex motion and/or details . I think the main difference was the source - the one you posted looked to be low in detail (it looked like a multi generation encode, oversmoothed), so the resulting differences were fairly minor to the human eye (you'd still be able to measure the differences with metrics). For the same reasons, you wouldn't see large differences on "simple" cartoons either, but you would see it to a larger extent on modern high detail CG

    You can repeat the tests with --preset veryfast, which has mb-tree, and you will still see differences. The point is that there are differences in quality, both measureable and visible at the same CRF
    Quote Quote  



Similar Threads

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