Hacker News Reader

Comment by jwr | Hacker News Reader

Notice he's leaning towards 48p as opposed to 60p, which is an already standardized framerate that your modern TV likely supports.

I do not think that is an accident. After all, once we all buy our new 3D TVs, something will have to happen for people to buy new TVs again in 3-4 years.

And before you mention bandwidth considerations, I do not have actual data, but I suspect that when using H.264 High Profile, the difference in bandwidth between 48fps and 60fps won't be that large. There is a limit to how much you can change visually in a single second: human perception.

View on HN

48 divides evenly into 240, so in theory a 240Hz TV could take a 48Hz signal and display it without any pulldowns, while 120Hz is 2.5 times 48, so it needs to do some work to create extra frames. (And really, that's the main advantage, currently, of 120Hz and 240Hz: 24, 30, and 60 (the primary American frame rates) all divide evenly into them, unlike 24 into 60).

Seems like you wouldn't count on the TV to do this work. Instead, you'd shoot at 48, then convert to 60 before delivery, in the same way that TV shows shot on film and edited at 24p subsequently convert to 25 for PAL markets and 29.97 for NTSC.

I would rather count on the media player (Blu-ray player or whatever) or TV to do it than in the source. Think of it this way: in NTSC, DVDs were either had hard 3:2 pull down for 24Hz to the 29.97Hz to show on the interlaced CRTs of the time. But now that we have TVs that can show 24Hz, if you play those DVDs they still have the interpolated frame artifacts. While if you left it up to software pulldown in the DVD player (which they basically all had, since I believe it was part of the standard), they can now be shown at the proper frame rate without interpolation.

Most theatrical DVDs released after the early 2000s were 24Hz, as far as I know and depended on the DVD player to convert to NTSC. So now in modern players, they're outputted at 24Hz for 120Hz and 240Hz TVs.

The only downside is if you get crappy TVs and/or Bluray players that do horrible pulldown, I guess. But I've gotten hardpulled down DVDs that look horrible as well (I'm looking at you, BBC).

Problem is, while you can devise ways of converting 24p to interlaced 29.97, getting from 48 progressive frames to 60 progressive frames isn't obvious to me at all. You'd have to add a frame every 4 frames. This would likely ruin the whole large framerate experience.

Assuming the bitrates are the same, 48fps will require 80% of the bandwidth needed to handle 60.

To the best of my knowledge, H.264 only supports two GOP structures (short = 6 frames / long = 15 frames), both of which operate independently from the playback speed.

In other words, a short GOP setting produces 8 blocks per second running at 48, and 10 per second running at 60. Assuming a constant bitrate (I know, I know), the throughput of the entire system needs to increase by an additional 20% to handle 60. And that's where things fall apart.

Consider the iTunes store, where I believe HD rentals top out at 720p. Or most HD broadcasters, who skew away from the 19.39 Mbps ceiling on the ATSC MPEG-2 spec in favor of programs encoded at 13.something Mbps (the legal floor) thereby allowing an SD channel encoded at ~6Mbps to be delivered in parallel. Does this exercise in bandwidth maximization impact the quality of the HD component? You bet. Do the networks care? If an additional SD means they can dramatically increase the number of ads served, then no, absolutely not.

So regardless of how great the acquisition setup, and how awesome the playback system, there's always the network in between them to consider. And with providers moving toward bandwidth caps, I see picture quality going down, not up. So, streaming 1080p60 H.264 encoded at (say) 7-9 Mbps? Choice, yes, but total fantasy for the time being.

Of course, Cameron is talking about making theaters competitive, so maybe that's his point. Given that BluRay has effectively failed as a universal home standard, and that instant-availability is hugely favored over order-and-wait programs, image quality of filmed entertainment at home is now tied to network limits.

The real question, then, is "are there enough cinephiles who actually care?"

"Assuming the bitrates are the same, 48fps will require 80% of the bandwidth needed to handle 60."

Not exactly - while your 80% approximation is true for uncompressed footage, h.264 encoded video and essentially every modern day video compression schema will inherit data from the frame(s) before it. The higher the framerate, the less (generally speaking) the images will change between frames - leading to a reduction in data per frame.

Here's an example:

File 1: http://files.jjcm.org/60fps.mp4 - 127KB

File 2: http://files.jjcm.org/48fps.mp4 - 119KB

As seen here, the 48fps video 94% of the size of the 60 fps video. Granted, this is a singular example, but at least it supports the concepts that I described above.

Wow, you're absolutely right about the (minor) file size differences. That's really interesting. H.264 FTW.

I have to admit, I couldn't see any difference in the quality of the motion between the two clips, but thanks for posting them.

H264 has no such limitation regarding gop structure. The levels (3.0, 4.1 etc) define (slightly indirectly) the DPB (decode picture buffer) size. This is how many frames must be held in memory at one time. The gop itself is not limited. Furthermore the compression rate for high frame rate actually increases for live action as there has been less change between the frames and even taking a fixed framerate. There are compromises to be made and faster framerate with a fixed bitrate may not be great for all usages but I think that you are being very, and unduly, pessimistic here. From my experiences at 720p 60fps and a 6 Mbps stream (which we currently have deployed) and a nice encoder you are looking at a very nice picture.

Very interesting. Thanks for the detail. I guess the fixed GOP structure must be limited to HDV (MPEG-4). That flexibility is pretty awesome.

For what it's worth, I suspect what you describe (720p60 @ 6Mbps) is more than fine for most people. In other words, Cameron's idea that people will be going to theaters because of magical picture quality seems like wishful thinking. I suspect people go to theaters because they like going to theaters. It's social. It's big. It's immersive and (relatively) distraction-free - all part of a fun night out.

Are they going to go home to 720p60 @ 6Mbps and think 'gah!'. Unlikely.

HDV has nothing to do with H.264, or any of the MPEG-4 standards for that matter.

You're absolutely right. I see that HDV is encoded with the H.262/MPEG-2 Part 2 compression scheme.

> "To the best of my knowledge, H.264 only supports two GOP structures (short = 6 frames / long = 15 frames), both of which operate independently from the playback speed. In other words, a short GOP setting produces 8 blocks per second running at 48, and 10 per second running at 60. Assuming a constant bitrate (I know, I know), the throughput of the entire system needs to increase by an additional 20% to handle 60."

I'm sorry, but it needs to be stated: the above isn't true and the calculations have nothing to do with reality. H.264 supports pretty much any GOP structure and you don't get from frames to final bitrate by dividing frame rates by GOP size.

Dude, chill. Also, read.

Accepting that I was mistaken about the rigidity of h.264 (a possibility I acknowledged up front), and assuming that you DID establish a GOP structure of 6 frames (at a constant bit rate) shooting for one minute at 48 fps would produce 80% of the data generated by shooting at 60fps.

Indeed, this ratio would be true if you used no compression at all. The point was that in a fixed GOP scheme (again, at a constant bitrate), the difference in FILE SIZE would increase with the difference in frame rate, so the idea that file sizes for 60 fps material would only be marginally larger than those for 48 fps seemed mistaken.

My only mistake was in assuming that the GOP is fixed in h.264. I agree that "you don't get from frames to final bitrate by dividing frame rates by GOP size." But then again, I never said you did.

So because one country is willfully going back to the stone age with caps, everone living in other less idiotic countries have to suffer?

Since no one else responded with this: you're right, it is not an accident that he would be considering 48fps, the reason being that a good amount of 24fps film projection occurs at 48fps on equipment that double pumps each frame. Such equipment (and the related equipment and processes around it) would likely be easier to upgrade/retrofit to 48fps than to 60fps.

I believe there are already some TVs that accept 48Hz as an input frame rate over HDMI. This is the only way 48fps video will look right, unless we move to 240Hz (the LCM of 48 and 60). 120fps would give the same juddery results as playing 24fps on a 60Hz display, but going up to 180 might be good enough to make the judder invisible (I don't notice the judder of a 24fps stream on an 85Hz CRT, but do at 60Hz, for example).

Reply on news.ycombinator.com