SMW 17: Beamr's Dror Gill Talks Integrated Video Encoding and Optimization

蒂姆Siglin: 欢迎回到流媒体西部2017. 我是蒂姆·西格林, 流媒体特约编辑, 同时也是ReelSolver的媒体策略负责人, 合并. 今天和我在一起的是dr Gill, Beamr的首席技术官. So, Dror, first of all, before we go into some of the topics, tell me a little bit about Beamr.

Dror吉尔: Beamr is the leading provider of video encoding and optimization solutions, 我们的主要客户是mso, 电信公司, 和OTT提供商. 我们提供基本的HEVC和H.264 encoding solutions, which are very high density and high quality. 我们还提供所谓的视频优化解决方案, which are pieces of software that take existing encoded video files and make the bit rate lower without hurting their quality.

蒂姆Siglin: 事实上是这样的, 我曾在一家名为EuclidIQ的公司工作过一段时间, 担任他们的策略主任, 所以从这个角度来看，我对Beamr非常熟悉. 你们去年收购了先锋视频?

Dror吉尔: 是的. 去年我们收购了先锋视频. We wanted to have tighter integration between the codec and the optimization software. 所以我们出去了, 我们考察了10家不同的公司, and we found out that Vanguard was the best video encoder out there. 我们于2016年3月收购了该公司, and since then we've been working very closely with them on integrating the encoding technology with the optimization technology.

蒂姆Siglin: So, that ability to go really deep down into the encoder and add the optimization in, at that level. 我从IBC听到了一些关于你们的好消息, that people were saying that integration has actually yielded significant benefits, 无论是AVC还是HEVC.

Dror吉尔: 正确的. We've seen bit rate reductions of around 20 to 50%, typically, on content. 回到NAB, 我们发布了这个产品, 内置优化的Beamr 5倍HEVC编码器, 它一直做得很好.

蒂姆Siglin: And your basis, when you say 20 to 50%, that's against something like x265, or ... 你的底线是多少?

Dror吉尔: Yeah, this is when you compare it to any other encoder, working in a regular VBR mode. In VBR mode, 您设置了目标比特率, that's the average bit rate you will get. In our encoder, there's an additional mode called CABR, content adapted bit rate. 当你运行它的时候, 您设置了目标比特率, 但实际的平均值会更低, 这是基于内容的. 我们在帧级进行内容自适应编码. 对于每一帧, we analyze what type of content and how many bits we should allocate, and that's how we can get a very low bit rate on each one of the titles.

蒂姆Siglin: 正确的. In fact, I just had David Sayed from Brightcove, who was talking about context-aware encoding. Yuriy Reznik from InterDigital has joined them, and it seems like there's this ... 之间的, 你知道, 当我在欧几里得的时候, 与Beamr, 和尤里在做的事, 做视频优化确实是一种进步, 我想要这个已经十多年了, 我很高兴看到我们终于做到了.

Dror吉尔: 我也很高兴这件事终于发生了. 我们是第一个. 我们从2009年开始做视频优化. 我们从jpeg开始. 在2012年或2011年，我们从H开始.264年优化. 现在，整个行业终于明白了其中的含义. It doesn't make sense to encode all of your videos at the same bit rate, 因为每个视频都有不同的要求. 现在你看到Netflix也在这么做, 每个人都喜欢这个头衔, 按照场景优化.

蒂姆Siglin: I think we're sort of half way through the continuum, because we've moved to per-title. 我真的很想看到我们在每个场景, 或每张照片, 因为即使在同一个标题中, 我们有很多不同类型的内容. 所以如果你支持这一点，不仅仅是编解码器的未来, 而是编码, 那真的很有帮助.

Dror吉尔: 是的. And we're one step ahead of that, because we're already doing per-frame. 场景内部, 我们正在适应视频的每一帧, 我们从一开始就是这么做的. This is really how you can really get the optimal bit rate for each piece of content. 在某些情况下, we saw that customers sent us a reel of different types of content concatenated together. If you have a naïve method that just looks at the beginning and tries to estimate what it is, 它会失败的. 但是如果你看每一帧, 并尝试找到正确的编码参数和比特率, 这样你才能真正适应不断变化的环境.

蒂姆Siglin: 显然，你还需要帧间的好处 ...

Dror吉尔: 完全.

蒂姆Siglin: ... but if you can do intra-frame, 如你所说, that's your starting basis.

Dror吉尔: 正确的. 你必须看, at least at pairs of frames because preserving the temporal flow is very important.

蒂姆Siglin: 正确的. 对，空间和时间，你得同时处理. So the two panels you’re on today--you said you moderated one on HEVC implementations. 然后是视频编解码器的未来. 我想有人把它描述为HEVC-AV1的重击. 你对HEVC, AV1有什么看法? 显然，你拥有一家拥有HEVC压缩的公司.

Dror吉尔: 正确的. I was in the audience of a third panel that also talked about HEVC vs. AV1. 在那个小组里，他们之间更像是一场战斗. 在我参加的小组讨论中, it was more about encoding techniques and streaming and things like that. 我们从一开始就非常相信HEVC. And actually, the company we acquired, Vanguard, has been developing HEVC for the past five years. 也是最大的——我认为是最大的——OTT提供商之一, 这是众所周知的, 他们所有的4K HDR系列都使用我们的编解码器.

We continue to push forward in HEVC because we believe that this is the way to go in terms of standard. 比特率效率在那里，质量在那里. 市场的发展不如我们想象的那么快, 这主要是由于版税的不确定性.

蒂姆Siglin: 这是版税的不确定性, 但我也感觉到, 特别是上下文感知编码, AVC已经足够好了, 除了超高清和4K. It's good enough, and if you can get some of that extra benefit, the workflow stays the same.

Dror吉尔: When we were pushing content-adaptive encoding, we were saying, you can reduce 20 to 50% of your H.264比特率, 所以你不需要去HEVC, 现在你有HEVC了, 我们知道HEVC可以节省30美元, 40%, 而是内容感知HEVC, 就像我们的beam 5x产品一样, 你可以在此基础上再节省30%到40%. That’s because--and this is something that came up in the panel that I moderated--with HEVC, 确实没有任何感性的工具来定义这个标准. 所以这是可以实现的. The efficiency you get in HEVC is because of better mathematical models, 压缩块大小之类的东西. But the basic perceptual principle of optimizing video and adapting it to the content, 对HEVC和H同样有效.264.

蒂姆Siglin: 因为, 如你所说, there was nothing added into the standard that was different from what was in AVC.

Dror吉尔: 完全. 这就是为什么我们的内容自适应也适用于HEVC. 现在我们正在推动HEVC，因为我们想说, 好吧, 你可以得到双重好处, HEVC和内容自适应.

蒂姆Siglin: 好吧, 听起来很有趣, 因为我想知道, 我们经历了AVC减少的一段时间, HEVC是否会出现类似的情况, 听到你说你得到了HEVC的双重效果, 以及在此基础上的优化, 这是一个非常有说服力的论点.

Dror吉尔: 我甚至会说，今天有了HEVC，我们可以节省更多的钱. 原因是H.264年实现, 比如开源的x264, 经过多年的优化和发展了吗, 并且达到了很高的水平. 对于HEVC，实现才刚刚开始

蒂姆Siglin: 对，x265就是一个例子.

Dror吉尔: 正确的. 它落后了，所以你还可以存更多钱 ... 今天, 还有更多的冗余, 因为编解码器本身, 实现, 没有被优化. 但是，即使在最优的H上.264, we can save a lot, and I believe this will continue with HEVC. So the debate of which codec to use, I think maybe last year, at this show, it was still open. 但今年4月，三星宣布了他们的支持. 今年6月，苹果宣布了他们的支持. 现在iOS 11发布了. 到今年年底, we'll have hundreds of millions of devices supporting HEVC in the field, 当然是所有的4K电视, 微软的Edge浏览器. If you have HEVC support in the Skylake chip on your PC, supports HEVC. 我相信明年, 2018, will be the year that we'll start to see massive deployment of HEVC in the field, 到电视机, 移动设备, 所有的OTT流媒体服务. We're already starting to see demand from our customers, more demand for the HEVC codec.

蒂姆Siglin: 最后再想一下:这就是原因, 从AV1的角度来看, 对他们来说，得到他们所拥有的东西是如此的有吸引力, 进入市场, 在HEVC成为事实之前. 如果你站在他们的立场上思考. They have a window of time in which they can go out and argue against it, but 优化部分 that you're putting in there makes the argument even more difficult.

Dror吉尔: 正确的. 也, 如果我们谈到编解码器的成熟度, HEVC, 软件实现, 我们在过去的五年里一直在发展. AV1会冻结, 正如我们在节目中听到的, 在接下来的一两个星期里, 但它是一个参考实现, 不是最优的. So, 确定, 您可以在云上启动多达100个服务器并执行一个流, 这是一个很好的演示, 但明天就这么做是不现实的, 或者在2018年, 用于直播.

蒂姆Siglin: 好吧. 好吧, 我要在这两者之间保持中立, 但很高兴接到你的电话, 优化部分, 在HEVC中有比AVC更多的益处. 我真的很感谢你的时间.

Dror吉尔: 非常感谢.

蒂姆Siglin: 绝对引人入胜.