4 ms·

From Cinepak to H.265: a brief history of video compression

27 points by e1ven 17y ago

ars 17y ago
It's too bad ATSC didn't wait for H.264 to be practical. You could fit 4 to 8 channels in the space of one MPEG2.
- BigZaphod 17y ago
  To make progress, at some point you have to stop waiting for the next new thing and use what you've got right now instead. (See the Duke Nukem story: http://www.wired.com/magazine/2009/12/fail_duke_nukem/ http://www.wired.com/magazine/2009/12/fail_duke_nukem/)
  - andrewvc 17y ago
    Agreed. We've just started working with Canon 5D 1080p h264 video files in Premiere. Even on a new Core i7 machine, with RAID 0 disks, Premiere just doesn't work very smoothly with h264 video (we convert it to Cineform Aspect HD as an intermediate). MPEG2 by comparison, is lighting fast. If you go back to when HDV was new, people were bitching about how slow MPEG2 was compared to DV (since it wasn't all I frames). That's just the way progress works.
DarkShikari 17y ago
Overall a pretty good, albeit quite brief, article, though the description of H.265 is terrible: H.265 will be lossier than H.264, in other words, but lossy in ways that won't be too obvious to humans and our imperfect image-processing brains. Pause an H.265 video and break out the spyglass, and you'll find many technical imperfections compared to older codecs, but it's all about perceived quality when the moving picture is, well, moving. The way in which loss is applied to the video is generally determined by the encoder, not the video format. The Joint Video Team is still busy optimizing everything for PSNR, so they don't appear as if they're going to change the entire concept behind video coding just yet. For example, we'd benefit hugely from coding-layer improvements to increase the visual energy of the video, much as Vorbis's failure mode is noisy audio as opposed to metallic warbling. H.265, so far, is shaping up to be larger macroblocks, a few much-wanted improvements such as a larger transform size, and a huge amount of brainstorming that has resulted in an enormous number of really stupid ideas. I'd give it at least another half decade before it has a chance of becoming a real spec, if only to filter out the bad ideas to find the really good ones. There's tons of interesting stuff there though, and it's worth following if only to read the JVT [Joint Video Team] papers. The real problem though is that H.265 has no chance of possibly satisfying the needs of those who really want it. The cable companies really really want to be able to solve the following problems in one fell swoop: 1. They want to have hundreds of channels, but can't bother laying new wire, so they want each HD channel to fit in an SD mux (~3.25 megabit CBR). 2. They want to get rid of interlacing, but they also want the smoothness of motion that 60 fields per second gets, and they also want the marketing name of 1080, so they want to switch to 1080p60. From what I've heard so far, H.265 won't even support interlacing, which is the single best decision I've seen out of the JVT. 3. They want this all available in reasonably-priced implementations as soon as possible. Oh, and it can't have significantly higher hardware requirements for decoding either. By comparison, x264, one of the best H.264 encoders, on its absolute slowest, non-realtime settings, can make 1080p30 look tolerable at about 3-4mbps. In other words, they want a 2-4x improvement over H.264, for zero extra CPU cost. They want fucking unicorns. This isn't going to happen.
- gcb 17y ago
  thanks. i've never understood why we had xvid, divx, quicktime6 et all sice they are the same format. and the article failed miserably on this topic. but your line about the encoder being responsible for the lossles made me see it. although i think it would be irresponsible to claim that the lossles possibilities a format allows for the encoders don't play any part in it.
mdasen 17y ago
I'm really hoping that Google opens VP8 up. From what I've seen, VP8 provides better image quality while requiring less CPU power to decode. If Google hopes that the <video> tag is going to take off, we'll need a freely available video codec that also has a company behind it that can sell patent guarantees. Firefox can't redistribute a non-free codec and Apple and Microsoft won't distribute something that leaves them open to potentially huge liability.

27 points by e1ven 17y ago

ars 17y ago

It's too bad ATSC didn't wait for H.264 to be practical. You could fit 4 to 8 channels in the space of one MPEG2.

BigZaphod 17y ago
To make progress, at some point you have to stop waiting for the next new thing and use what you've got right now instead. (See the Duke Nukem story: http://www.wired.com/magazine/2009/12/fail_duke_nukem/ http://www.wired.com/magazine/2009/12/fail_duke_nukem/)
- andrewvc 17y ago
  Agreed. We've just started working with Canon 5D 1080p h264 video files in Premiere. Even on a new Core i7 machine, with RAID 0 disks, Premiere just doesn't work very smoothly with h264 video (we convert it to Cineform Aspect HD as an intermediate). MPEG2 by comparison, is lighting fast. If you go back to when HDV was new, people were bitching about how slow MPEG2 was compared to DV (since it wasn't all I frames). That's just the way progress works.

DarkShikari 17y ago

Overall a pretty good, albeit quite brief, article, though the description of H.265 is terrible: H.265 will be lossier than H.264, in other words, but lossy in ways that won't be too obvious to humans and our imperfect image-processing brains. Pause an H.265 video and break out the spyglass, and you'll find many technical imperfections compared to older codecs, but it's all about perceived quality when the moving picture is, well, moving. The way in which loss is applied to the video is generally determined by the encoder, not the video format. The Joint Video Team is still busy optimizing everything for PSNR, so they don't appear as if they're going to change the entire concept behind video coding just yet. For example, we'd benefit hugely from coding-layer improvements to increase the visual energy of the video, much as Vorbis's failure mode is noisy audio as opposed to metallic warbling. H.265, so far, is shaping up to be larger macroblocks, a few much-wanted improvements such as a larger transform size, and a huge amount of brainstorming that has resulted in an enormous number of really stupid ideas. I'd give it at least another half decade before it has a chance of becoming a real spec, if only to filter out the bad ideas to find the really good ones. There's tons of interesting stuff there though, and it's worth following if only to read the JVT [Joint Video Team] papers. The real problem though is that H.265 has no chance of possibly satisfying the needs of those who really want it. The cable companies really really want to be able to solve the following problems in one fell swoop: 1. They want to have hundreds of channels, but can't bother laying new wire, so they want each HD channel to fit in an SD mux (~3.25 megabit CBR). 2. They want to get rid of interlacing, but they also want the smoothness of motion that 60 fields per second gets, and they also want the marketing name of 1080, so they want to switch to 1080p60. From what I've heard so far, H.265 won't even support interlacing, which is the single best decision I've seen out of the JVT. 3. They want this all available in reasonably-priced implementations as soon as possible. Oh, and it can't have significantly higher hardware requirements for decoding either. By comparison, x264, one of the best H.264 encoders, on its absolute slowest, non-realtime settings, can make 1080p30 look tolerable at about 3-4mbps. In other words, they want a 2-4x improvement over H.264, for zero extra CPU cost. They want fucking unicorns. This isn't going to happen.

gcb 17y ago
thanks. i've never understood why we had xvid, divx, quicktime6 et all sice they are the same format. and the article failed miserably on this topic. but your line about the encoder being responsible for the lossles made me see it. although i think it would be irresponsible to claim that the lossles possibilities a format allows for the encoders don't play any part in it.

mdasen 17y ago

I'm really hoping that Google opens VP8 up. From what I've seen, VP8 provides better image quality while requiring less CPU power to decode. If Google hopes that the <video> tag is going to take off, we'll need a freely available video codec that also has a company behind it that can sell patent guarantees. Firefox can't redistribute a non-free codec and Apple and Microsoft won't distribute something that leaves them open to potentially huge liability.