Lower video resolution can deliver better quality, says Netflix Netflix has revealed a new plan for ensuring its videos arrive in your device looking their best, which can sometimes mean streaming them in lower resolution. The streamer calls its new scheme “Per-Title Encode Optimization”. As the name implies, the company is now analysing the content it offers and delivering it in different …
I may have missed something but it sounds like Netlix is trying to match the screen size and maximum resolution of the stream. In photography, size of the photo is directly proportional to the maximum size it can be printed before artifacts become noticeable and annoying. Not strong on video resolution and screen sizes.
They don't use enough bandwidth to support simultaneous high res and high detail or motion video. They've taken the cheap option of trading reduced spatial resolution for enhanced motion resolution (and simply thrown away detail), cheap because it saves them money on bandwidth, storage and processing, 20% storage reduction has been suggested. Now they need to convince punters it's all for their benefit, not just Netflix profits.
Unfortunately they can't really rely on increasing bandwidth, the US infrastructure isn't good enough, so customers just get a different bandwidth caused problem to suffer. Given how many seem unable to spot upscaled video it's probably the best choice.
Can't be as bad as iPlayer's dynamic resolution scaling, which is extremely noticeable and highly irritating. More accurately it can't be if they want to stay in business.
Pixel resolution is a fairly meaningless quality measurement in video anyway.
If the video is 3840x2160 but breaks up into clearly visible encoding blocks at any point, the video is unwatchably bad.
Even if it doesn't break up, the sustained bitrate you can actually get from your ISP will probably wipe out any possible extra detail.
At 1280x720 the same bandwidth can probably encode a much better looking video, and your 4k screen can upscale the result to give something objectively and subjectively better.
It's not Netflix that created the demand for 4k though, is it? They probably did know this all along, but idiots buying 4k TVs have demanded 4k content and have been happy to have content that looks worse than 1080 because it says it has more pixels. If TV companies had developed better smart systems rather than cramming in more pixels we'd be better off overall. As it is, the smart interface in most 4k TVs is already obsolete so they won't have access to Netflix for long anyway...
@Neoc - they would be absolutely insane to encode the streams live, and I hardly need to read the article to tell you that the don't do that. Encoding the same video over and over each time somebody watches it makes no sense whatsoever. But allow me to quote from the Netflix tech blog post that this article refers to, just to be absolutely clear;
"We pre-encode streams at various bitrates applying optimized encoding recipes. On the member’s device, the Netflix client runs adaptive streaming algorithms which instantaneously select the best encode to maximize video quality while avoiding playback interruptions due to rebuffers."
@tony72
The Reg does seem to be cherry picking what it wants to report. No mention of the offline re-encoding project that aims to drop storage by 20% for those master copies or any guesswork about what that recoding actually does (I'd guess they're just turning on constant Q and generating variable bitrate but that's just a guess). That recoding will be powered by their cloud instances in spare time.
The Reg concentrated exclusively on the lowered resolution scheme and guesswork about some improbable and unnecessary real time transcoding.
some improbable and unnecessary real time transcoding.
Certainly improbable in this instance - because it is unnecessary.
A few years ago, I worked on a device that offered[1] up to 144 simultaneous real-time transcodes. The idea was that multiple device types combined with NVOD would mean such things were useful - and as this all fit in a 1U case, that was quite a transcode density. But you don't do stuff like that in software, even if you do like Docker...
I'm not sure if the company sold any, mind.
Vic.
[1] I found out a few weeks ago that the current spec is for 72 simultaneous transcodes. Ho well...
Including relatively programmable silicon that can be configured to handle a variety of source and target encodings.
Well, I worked for the market leader in this field. And they used FPGA.
Now I'm not going to tell you that this is the only way to do it - I haven't done sufficient research for that - but I can tell you that this is how it is done, despite the fact that the people choosing to do this had carte blanche in terms of how they did it (including generating custom silicon).
Vic.
They've taken the cheap option of trading reduced spatial resolution for enhanced motion resolution (and simply thrown away detail)
So has everyone else.
Digital TV is lossily compressed. You've got a tradeoff between resolution, frame rate, ability to cope with motion[1], and bandwidth used. Different providers pick different compromises.
What Netflix is saying is that, for some types of content, the artefacts caused by dropping the transmitted resolution are less noticeable that the artefacts introduced by encoding a higher resolution at lower quality. And they are exactly right; the trick is in determining what sort of content you've got...
Unfortunately they can't really rely on increasing bandwidth, the US infrastructure isn't good enough
No infrastructure is good enough; given sufficient framerate, frame size, and quality, any infrastructure will be overloaded eventually.
Disclosure: I've been in this industry for quite a while...
Vic.
[1] Simple Y- and Y- translations compress well; this is what MPEG expects. Zooming and rotating are far more costly in terms of bandwidth used. I've found the old Doctor Who title sequence to be quite exacting.
I would sooner have analogue TV with ghosting on the picture rather than having super hi-def video that buffers every 5 minutes.
Kids today won't know that something that's lower quality but plays uninterrupted is always going to be 100% better than something that's high quality but has to buffer every 10 minutes.
I'll stick with my native 720p/1080i/1080p24 Panasonic Plasma for a while longer then... eventually there will be a reasonable amount of 4K/UHD content with a reasonable balance between compression and image quality but for now it's a buzz word designed to entice the vulnerable who 'need' the latest TV - even if they don't quite understand why.
"These days, the company's decided that approach was a bit arbitrary because it can result in artefacts appearing during busy moments of a complex film, while also using rather more resources than were required to stream something simpler, wasting storage and network resources along the way."
So they haven't heard of ffmpeg -crf 18 ?
Constant Rate Factor (CRF) does exactly what is described above, in that it compresses down the minimum size for the selected level of visual quality. Granted, this means the bit rate isn't fixed/constant, but given that some buffering will be happening anyway that isn't that big a problem most of the time.
The catch is you can't assume unlimited buffering capacity, so the encoding rate may still be constrained by a relatively short moving window. On something like a smart TV that could be a low limit, far too little for a 10min action sequence for instance. At that point only reducing resolution or framerate will fulfil the buffering, encoding bandwidth & artefact constraints.
It's a bit more than that but it's related. They calculate, "At each QP point, for every title, ...the resulting bitrate in kbps, ...and PSNR (Peak Signal-To-Noise Ratio) in dB, "...and from that they calculate, for a given bandwidth, which resolution is best for a particular movie.
Constant Rate Factor (CRF) does exactly what is described above, in that it compresses down the minimum size for the selected level of visual quality
That's probably unusable in most broadcast contexts. You don't have the bit budget to just allocate more bandwidth when needed; even VBR has min & max settings, because otherwise, something downstream gets swamped. And that means breakup at the client end, which doesn't make you popular.
Vic.
We've had FTTH for about a year and a half now. We have selected the 150 Mbps service, not yet taking their more expensive nearly-Gb offering. Several family can be watching several HD videos from YouTube at once, no issue. Certainly 'ready for prime time'.
We don't subscribe to Netflix because their library leans towards 'brain pablum', tedious entertainment crap typically of no intellectual value. YouTube offers a vast selection of far more educational material.
@Dan Wilkie
If one goes outside and randomly eats something, it'll likely be dirt or a twig. However, if one can manage to find one's way from one's home to a fine restaurant, then chances are good that it'll be a lovely meal.
YouTube is better if you use the Search function to find something you're interested in. Don't just randomly select random videos.
I shouldn't have to explain this towards refuting your attempted point.
My Humax Freesat DVR finally got a Netflix app this week, so I signed up for a free month to the HD package (and immediately cancelled to avoid auto-payment when the free month is up, but you still get to use your free month). I couldn't understand why the picture wasn't sharp on my wired 40 Mbps fibre connection and 1080p plasma though.
It turns out Netflix defaults to "Auto" playing mode and the info button on the remote confirmed it was only playing HD content at 480p despite my decent setup. I had to go onto the Netflix Web site and flick the playing mode to "High" - as soon as I did that, the HD movie I was playing flicked from 480p to 1080p mid-stream, with no buffering or pixellation and a much sharper picture. In other words, Auto mode is terrible at estimating your bandwidth and I wonder how many HD subscribers have been playing back at 480p because of the crappy Auto mode?
Talking of HD, if the vast majority of Netflix content is available in HD and it's likely the vast majority of Netflix potential or paid-up subscribers have HD TVs, then why do Netflix charge an extra 1.50 per month for HD? Netflix will claim it's for the bandwidth usage (they muddy this by offering 2 simultaneous devices in the HD package), but the Auto default playing mode selecting 480p (at least for me) sounds like a convenient bug/feature to save using bandwidth on HD accounts to me....
I'm pretty sure that Netflix upscale whatever they're showing you to be full screen.
Now, instead of having a full resolution video which falls apart into encoding blocks as soon as something moves, you'll have a video which looks lower quality, but is consistent in that level of quality.
If only there was some way that one could pre-buffer an entire video beforehand, at high resolution. Rather than sending it over the internet, it might even make sense to ship the file on physical media instead on, say, a DVD?
Netflix should get right on that.
Some OTT operators in Europe have been doing this for a while.
The process requires dynamic adjustment of the ABR in the player or the ability to rewrite a manifest based on other users' player telemetry.
This is not new technology or a "massive game changer".
Netflix just has a better PR team.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
