Announcement

Collapse
No announcement yet.

Eve: A New VP9 Video Encoder Offering Much Better Results

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Eve: A New VP9 Video Encoder Offering Much Better Results

    Phoronix: Eve: A New VP9 Video Encoder Offering Much Better Results

    Eve is short for the Efficient Video Encoder and reportedly offers much better results than existing video encoders for Google's VP9 format...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Good, I guess. It's funny though that libvpx always get beaten by a third party.

    Comment


    • #3
      Nice, but:
      - the encoder doesn’t seem to be available
      - the tests with PSNR and SSIM are worthless since they do not correlate with visual quality (that’s what I read anyway)
      - the visual quality comparisons results can’t be trusted since the author may have picked specific frames that favour EVE (not to mention settings and bitrate)
      - note how he used --tune=psnr for the first tests and then removes it for the visual quality tests…
      - 5 times the encoding time of x264 for a similar quality is a hard sell…

      VP9 needs much better improvements to be competitive with x264. And in FOSS form, preferably. My hopes are on Daala/Thor (I wouldn’t mind hearing positive news about these ones!).

      Comment


      • #4
        "offers 15-20% better compression rates"

        what a ridiculous statement to make about a lossy codec

        Comment


        • #5
          Originally posted by xeekei View Post
          Good, I guess. It's funny though that libvpx always get beaten by a third party.
          What really concerns me is that x265 is just so much faster and easily offers the same quality as vp9, and the Alliance for Open Media chose libvpx/vp9 as starting point for their new codec. It is currently not very efficient and additionally very slow, I was hoping for a proper new codec

          Comment


          • #6
            Originally posted by utack View Post

            What really concerns me is that x265 is just so much faster and easily offers the same quality as vp9, and the Alliance for Open Media chose libvpx/vp9 as starting point for their new codec. It is currently not very efficient and additionally very slow, I was hoping for a proper new codec
            But isn't there some patent/royalty risk with x265? Certainly for HEVC h265. I always thought one of the main goals of vp9 was an attempt to get free of the whole patent royalty mess. But I haven't been following this for some time. Of course, our codec choices are pretty much driven by whatever Apple supports. And they are only going to endorse closed patent encumbered codecs.

            Comment


            • #7
              Originally posted by utack View Post

              What really concerns me is that x265 is just so much faster and easily offers the same quality as vp9, and the Alliance for Open Media chose libvpx/vp9 as starting point for their new codec. It is currently not very efficient and additionally very slow, I was hoping for a proper new codec
              *face palm*
              How could the Alliance for Open Media possible chose a proprietary codec as base for their free codec?

              Comment


              • #8
                Originally posted by atomsymbol

                What do you mean?
                Well, by itself, it is a pretty ridiculous statement...one usually asks "at what cost?" In the case of a lossy codec, it's usually quality, but with lossy and lossless codecs it can be either quality, speed, or both.

                Comment


                • #9
                  Originally posted by xeekei View Post
                  Good, I guess. It's funny though that libvpx always get beaten by a third party.
                  Well, libvpx is simply that bad.

                  Originally posted by stqn View Post
                  Nice, but:
                  - the encoder doesn’t seem to be available
                  - the tests with PSNR and SSIM are worthless since they do not correlate with visual quality (that’s what I read anyway)
                  - the visual quality comparisons results can’t be trusted since the author may have picked specific frames that favour EVE (not to mention settings and bitrate)
                  - note how he used --tune=psnr for the first tests and then removes it for the visual quality tests…
                  - 5 times the encoding time of x264 for a similar quality is a hard sell…

                  VP9 needs much better improvements to be competitive with x264. And in FOSS form, preferably. My hopes are on Daala/Thor (I wouldn’t mind hearing positive news about these ones!).
                  - And it won't be. This is a commercial endeavor. So there will be a commercial sdk, but no open source or even a publicly downloadable encoder executable.
                  - This.
                  - Ronald is a reputable person, he wouldn't cheat. Encoder settings would be nice, but he probably used common presets. He did mention the bitrate, 200kbps for the 360p comparison.
                  - That's the only thing that makes sense. If you're going to plot a PSNR graph, optimize for that. If you're doing a visual comparison, optimize for that. I wouldn't even bother with the graphs, but there are for sure those among Ronald's prospective clients that get off on pretty graphs, so he has to have those as well.
                  - Agree. But this is just the beginning it seems, so there's plenty of room to assembly-optimize the hell out of the encoder. Ronald knows his way around assembly, he's the reason ffvp9 is so much faster than libvpx at decoding.

                  Originally posted by utack View Post
                  What really concerns me is that x265 is just so much faster and easily offers the same quality as vp9, and the Alliance for Open Media chose libvpx/vp9 as starting point for their new codec. It is currently not very efficient and additionally very slow, I was hoping for a proper new codec
                  I'm hoping that there will be competing encoder implementations of the Alliance codec from the start. This way the crappy state of libvpx won't be relevant. The most important thing for the Alliance codec will be a proper, well written spec. VP9 didn't even have a spec until like a month ago.

                  Originally posted by cjcox View Post
                  But isn't there some patent/royalty risk with x265? Certainly for HEVC h265. I always thought one of the main goals of vp9 was an attempt to get free of the whole patent royalty mess. But I haven't been following this for some time. Of course, our codec choices are pretty much driven by whatever Apple supports. And they are only going to endorse closed patent encumbered codecs.
                  Yes, h265/hevc is patent encumbered and the licensing situation is a mess (two patent pools, and even those cover only about half of all patents). And yes, the reason for the Alliance is avoiding the patent mess. However, Ion't agree with you about Apple's importance here, not when pretty much everyone but them is in the Alliance. If it was just Google that's one thing, but there are other big names in the Alliance, most notably Microsoft. Also, streaming companies such as Amazon and Netflix.

                  Comment


                  • #10
                    Originally posted by tessio View Post

                    *face palm*
                    How could the Alliance for Open Media possible chose a proprietary codec as base for their free codec?
                    The format is open, royalty free, and uses the New-BSD license. What's the problem?

                    Comment

                    Working...
                    X