Author Topic: eevBLAB #16 - 4K Render Youtube Experiment  (Read 20396 times)

0 Members and 1 Guest are viewing this topic.

Offline EEVblog

  • Administrator
  • *****
  • Posts: 26841
  • Country: au
    • EEVblog
eevBLAB #16 - 4K Render Youtube Experiment
« on: October 08, 2015, 02:31:55 pm »
Dave is running a test upscaling his 1080p 50fps footage to 4K UHD (3840x2160) 30fps and uploading to Youtube. Rumour has it that Youtube is supposed to detect this 4K footage and give better quality converted video quality.
Can you see any difference between this video at 1080p and this other one uploaded at 1080p?


And if there is a quality difference, is it worth the extra file size, rendering, transcoding, upload, and youtube processing time?

Render times for the 5min video are:
4K = 17:29
1080p = 1:59

Transcoding (Handbrake) times:
4K = 6:53
1080p = 1:09

Filesizes:
4K = 1281M
1080p = 312M

 

Offline EEVblog

  • Administrator
  • *****
  • Posts: 26841
  • Country: au
    • EEVblog
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #1 on: October 08, 2015, 03:12:32 pm »
I hope I understand this correctly. You shoot 1080p. You upscale it to 4K. YouTube downscales it to 1080p and I get to watch it in 1080 again.  And it might look better.   :-+

Yes. And that is exactly what people are claiming.

Quote
I don't really think you should do this but wouldn't shooting in actual 4k make more sense. Ersatz 4k seems an even greater waste of bandwidth and rendering effort.

Only for added flexability in editing. In terms of Youtube 4K playback, there is no perceptible difference between a true 4K camera and a 1080p camera upscaled.
 

Offline Muttley Snickers

  • Supporter
  • ****
  • Posts: 1800
  • Country: au
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #2 on: October 08, 2015, 03:15:39 pm »
I played them both back a number of times and could not see much difference if at all on my high spec pc and monitor, yet to check what they look like on the bigger LG smart TV, mind you and as I have previously pointed out with most Youtube videos I now tend to automatically watch at 480 and prior at 320 and will only kick it up to 720 or 1080  if I need to see something with finer detail providing that it is available in higher resolution.

After a couple of strokes my eyesight isn't what it once was and "blurry vision","flickering lights" and "tunnel view" now as standard built in features, honestly for most content I don't really give a rats how well a video is done nor how much effort went into producing it, although the exception being people such as yourself, Shahriar and a couple of others where the content is the determining factor and more important to me than anything else, these informative content based videos get both my admiration and accordingly my greatest appreciation regardless of the resolution.

And the day you start calling for makeup is the day that I pull the pin....... :palm:

Many Thanks

Muttley


And here's a free gift for those that need to watch in high definition....DIY.... :palm:
« Last Edit: October 08, 2015, 03:44:28 pm by Muttley Snickers »
One smart cookie, better make that two for good measure.
 

Offline rs20

  • Super Contributor
  • ***
  • Posts: 2047
  • Country: au
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #3 on: October 08, 2015, 03:43:10 pm »
I haven't even looked at the videos, I don't trust experiments like this unless they're double blind!  :)  However, I used youtube-dl to list the video formats and actual video data sizes of the two videos, and you can see that the 1080p version of the 4K upload is actually a smaller file (86.4 MiB / 103.23 MiB*) than the straight 1080p upload (87.20 MiB / 115.83 MiB). So the idea that the 4K uploads get a special, higher bitrate seems false.

Wild speculation: the upscaling to 4K used an interpolating filter, and when it's downscaled back to 1080p, the footage is very slightly blurred, leading to slightly easier compression for the codec.

* (A / B) where A is the VP9 codec file, and B is the AVC codec file.


rs20@cave:~/eevblab/1080$ youtube-dl -F q95mfnmMdwA   # Native 1080p
...
248          webm       1920x1080  DASH video 2616k , vp9, 30fps, video only, 87.20MiB
137          mp4        1920x1080  DASH video 4322k , avc1.640028, 30fps, video only, 115.83MiB

...

rs20@cave:~/eevblab/1080$ youtube-dl -F 4SqryR3-8uE    # 1080p --> 4K --> 1080p
...
248          webm       1920x1080  DASH video 2533k , vp9, 30fps, video only, 86.41MiB
137          mp4        1920x1080  DASH video 4318k , avc1.640028, 30fps, video only, 103.23MiB

...


Code: [Select]
rs20@cave:~/eevblab/1080$ youtube-dl -F q95mfnmMdwA
[youtube] q95mfnmMdwA: Downloading webpage
[youtube] q95mfnmMdwA: Downloading video info webpage
[youtube] q95mfnmMdwA: Extracting video information
[youtube] q95mfnmMdwA: Downloading DASH manifest
[youtube] q95mfnmMdwA: Downloading DASH manifest
[info] Available formats for q95mfnmMdwA:
format code  extension  resolution note
249          webm       audio only DASH audio   51k , opus @ 50k, 1.82MiB
171          webm       audio only DASH audio  111k , vorbis@128k (44100Hz), 3.88MiB
250          webm       audio only DASH audio  121k , opus @ 70k, 2.58MiB
140          m4a        audio only DASH audio  128k , m4a_dash container, aac  @128k (44100Hz), 4.55MiB
251          webm       audio only DASH audio  231k , opus @160k, 5.06MiB
160          mp4        256x144    DASH video  111k , avc1.42c00c, 15fps, video only, 3.90MiB
278          webm       256x144    DASH video  126k , webm container, vp9, 15fps, video only, 3.50MiB
133          mp4        426x240    DASH video  248k , avc1.4d4015, 30fps, video only, 8.75MiB
242          webm       426x240    DASH video  250k , vp9, 30fps, video only, 6.84MiB
243          webm       640x360    DASH video  434k , vp9, 30fps, video only, 12.68MiB
134          mp4        640x360    DASH video  610k , avc1.4d401e, 30fps, video only, 13.19MiB
244          webm       854x480    DASH video  747k , vp9, 30fps, video only, 22.53MiB
135          mp4        854x480    DASH video 1113k , avc1.4d401f, 30fps, video only, 26.15MiB
247          webm       1280x720   DASH video 1499k , vp9, 30fps, video only, 48.48MiB
136          mp4        1280x720   DASH video 2063k , avc1.4d401f, 30fps, video only, 50.84MiB
248          webm       1920x1080  DASH video 2616k , vp9, 30fps, video only, 87.20MiB
137          mp4        1920x1080  DASH video 4322k , avc1.640028, 30fps, video only, 115.83MiB
17           3gp        176x144    small ,  mp4a.40.2, mp4v.20.3
36           3gp        320x240    small ,  mp4a.40.2, mp4v.20.3
5            flv        400x240    small
43           webm       640x360    medium ,  vorbis, vp8.0
18           mp4        640x360    medium ,  mp4a.40.2, avc1.42001E
22           mp4        1280x720   hd720 ,  mp4a.40.2, avc1.64001F (best)

rs20@cave:~/eevblab/1080$ youtube-dl -F 4SqryR3-8uE
[youtube] 4SqryR3-8uE: Downloading webpage
[youtube] 4SqryR3-8uE: Downloading video info webpage
[youtube] 4SqryR3-8uE: Extracting video information
[youtube] 4SqryR3-8uE: Downloading DASH manifest
[youtube] 4SqryR3-8uE: Downloading DASH manifest
[info] Available formats for 4SqryR3-8uE:
format code  extension  resolution note
249          webm       audio only DASH audio   51k , opus @ 50k, 1.82MiB
171          webm       audio only DASH audio  111k , vorbis@128k (44100Hz), 3.88MiB
250          webm       audio only DASH audio  123k , opus @ 70k, 2.58MiB
140          m4a        audio only DASH audio  128k , m4a_dash container, aac  @128k (44100Hz), 4.55MiB
251          webm       audio only DASH audio  237k , opus @160k, 5.06MiB
160          mp4        256x144    DASH video  111k , avc1.42c00c, 15fps, video only, 3.90MiB
278          webm       256x144    DASH video  135k , webm container, vp9, 15fps, video only, 3.50MiB
133          mp4        426x240    DASH video  249k , avc1.4d4015, 30fps, video only, 8.75MiB
242          webm       426x240    DASH video  251k , vp9, 30fps, video only, 6.89MiB
243          webm       640x360    DASH video  439k , vp9, 30fps, video only, 12.75MiB
134          mp4        640x360    DASH video  610k , avc1.4d401e, 30fps, video only, 13.27MiB
244          webm       854x480    DASH video  738k , vp9, 30fps, video only, 22.37MiB
135          mp4        854x480    DASH video 1110k , avc1.4d401f, 30fps, video only, 26.18MiB
247          webm       1280x720   DASH video 1506k , vp9, 30fps, video only, 47.10MiB
136          mp4        1280x720   DASH video 2043k , avc1.4d401f, 30fps, video only, 49.84MiB
248          webm       1920x1080  DASH video 2533k , vp9, 30fps, video only, 86.41MiB
137          mp4        1920x1080  DASH video 4318k , avc1.640028, 30fps, video only, 103.23MiB
271          webm       2560x1440  DASH video 8050k , vp9, 30fps, video only, 224.87MiB
264          mp4        2560x1440  DASH video 8739k , avc1.640032, 30fps, video only, 239.34MiB
313          webm       3840x2160  DASH video 19787k , vp9, 30fps, video only, 660.01MiB
266          mp4        3840x2160  DASH video 22117k , h264, 30fps, video only, 695.09MiB
17           3gp        176x144    small ,  mp4a.40.2, mp4v.20.3
36           3gp        320x240    small ,  mp4a.40.2, mp4v.20.3
5            flv        400x240    small
43           webm       640x360    medium ,  vorbis, vp8.0
18           mp4        640x360    medium ,  mp4a.40.2, avc1.42001E
22           mp4        1280x720   hd720 ,  mp4a.40.2, avc1.64001F (best)


I hope I understand this correctly. You shoot 1080p. You upscale it to 4K. YouTube downscales it to 1080p and I get to watch it in 1080 again.  And it might look better.   :-+

That makes perfect sense if YouTube has code on its servers that chooses a higher 1080p bitrate if the source media is 4K. But as indicated above, that doesn't appear to be true, at least in this case.
« Last Edit: October 08, 2015, 03:45:49 pm by rs20 »
 

Offline Armxnian

  • Regular Contributor
  • *
  • Posts: 214
  • Country: us
  • Computer Engineering Student
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #4 on: October 08, 2015, 04:11:40 pm »
YouTube downscales it to 1080p and I get to watch it in 1080 again.
Youtube doesn't downscale to 1080p, it transcodes to 1080p. I'm pretty sure you're supposed to watch the 4k version, even if you're not on a 4k monitor (which will be downscaled), to get the extra bitrate. Someone needs to download the videos and compare the bitrate of the 1080p version of the 4k upload to the 1080p version of the 1080p upload. One thing to consider is that when uspscaling during rendering, the 4k version will require more bitrate to not look like crap, so you need to subtract that bitrate overhead from the extra bitrate that youtube gives you on a 4k upload. 4k is 4x the data of 1080, but I am not a video expert and have no idea how x264 or whatever encoder you use handles compression of 1080 vs 4k. Is a compressed 4k test file 4x the file size/bitrate of a compressed 1080p file? I don't think so. It can be less, or even more, based on the content and compression. I am also not sure if 4k of actual pixels take up more bitrate than 4k pixels that were produced by an upscaling algorithm. It's very complicated to quantify the quality of such a test objectively. However, with some testing and math, you can come up with an estimate.

I cannot comment on the subjective quality as my monitor fell off my desk, so I'm without a display/computer  |O

There are however other factors to consider. The hw acceleration implementation in chrome,firefox,etc isn't very good. Neither is software rendering. The people complaining about stutters in the comments need to realize that 4k playback requires a decent internet connection and a decent cpu, especially since the medium is a web browser.

One benefit with higher resolution sources is watching them on high resolution displays like many modern smartphones. The upscaling (1080->1440) in the youtube app and through a web browser isn't very good. I notice an immediate difference in quality watching a 1440 video on the native 1440 screen of my Nexus 6 compared to a 1080 source, even if the actual visual quality of the content is the same. One must appreciate the fact that the quality youtube delivers is amazing considering how much data Google has to store, process and deliver. But you must also acknowledge obvious limitations from an enthusiasts point of view.

I'm getting a professional 4k display on Tuesday so I can actually test everything.
 

Offline EEVblog

  • Administrator
  • *****
  • Posts: 26841
  • Country: au
    • EEVblog
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #5 on: October 08, 2015, 04:48:53 pm »
I haven't even looked at the videos, I don't trust experiments like this unless they're double blind!  :)  However, I used youtube-dl to list the video formats and actual video data sizes of the two videos, and you can see that the 1080p version of the 4K upload is actually a smaller file (86.4 MiB / 103.23 MiB*) than the straight 1080p upload (87.20 MiB / 115.83 MiB). So the idea that the 4K uploads get a special, higher bitrate seems false.

Someone on Youtube pointed out those same numbers.
 

Offline Howardlong

  • Super Contributor
  • ***
  • Posts: 4370
  • Country: gb
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #6 on: October 08, 2015, 05:35:31 pm »
I did a couple of native 4k videos on Youtube a few months ago to see how it worked out.

In the end, the massive increase in transcoding time at my end meant that I really didn't consider it worthwhile doing longer term for now. In addition, I only have facilities for 25 fps at 4k, I don't know if Youtube can cope with 50 or 60 fps on 4k: 1080p at 50 or 60 fps seemed a reasonable enough compromise.
 

Online Towger

  • Super Contributor
  • ***
  • Posts: 1356
  • Country: ie
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #7 on: October 08, 2015, 05:54:38 pm »
What. Perform two extra conversations on a lossy file format and expect better quality.  Smells like a Batteriser...
 

Offline rs20

  • Super Contributor
  • ***
  • Posts: 2047
  • Country: au
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #8 on: October 08, 2015, 06:11:44 pm »
What. Perform two extra conversations on a lossy file format and expect better quality.  Smells like a Batteriser...

 :palm: Two conversions with high quality settings can give a better result than a single conversion with low quality settings.
« Last Edit: October 08, 2015, 06:15:55 pm by rs20 »
 

Offline hans

  • Frequent Contributor
  • **
  • Posts: 978
  • Country: nl
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #9 on: October 08, 2015, 06:40:22 pm »
If you look careful at the white board drawings you can see some difference between 4K upload played back at 1080p, and the "native" 1080p upload.

The red text seems to be "jpegy" a bit. I uploaded a comparison picture zoomed in X2 below.
I took the comparison from frame 0 of each video in Firefox 41.0.1 & HTML5 YT player.

So if I did the comparison right, IMHO the 4K upscaled version looks better.. but also a bit silly that youtube works this way.
 

Offline EEVblog

  • Administrator
  • *****
  • Posts: 26841
  • Country: au
    • EEVblog
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #10 on: October 08, 2015, 06:59:18 pm »
So if I did the comparison right, IMHO the 4K upscaled version looks better.. but also a bit silly that youtube works this way.

Thanks for that. Yes, the 4K one does appear sharper.
 

Offline EEVblog

  • Administrator
  • *****
  • Posts: 26841
  • Country: au
    • EEVblog
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #11 on: October 08, 2015, 07:02:17 pm »
In the end, the massive increase in transcoding time at my end meant that I really didn't consider it worthwhile doing longer term for now.

As you can see with my numbers above, it is indeed a massive increase going to 4K in render time, transcoding time, upload time, and youtube processing time.
That's at 30fps of course. 50fps like I've been doing recently is also massively slower than 30fps.
I'm thinking about switching back to 30fps (but still shoot at 50fps) and only render and upload 50fps on videos that have a lot of movement, like outdoors stuff, walking cam footage etc.
 

Offline SteveLy

  • Regular Contributor
  • *
  • Posts: 220
  • Country: au
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #12 on: October 08, 2015, 07:04:36 pm »
It's Dave in his electronics lab not a bloomin Lucasfilm Ridley Martin Scott Scorsese billion dollar blockbuster. 1080p/720p is plenty good enough. Anything more is waste of time & resources.
 

Online coppice

  • Super Contributor
  • ***
  • Posts: 3675
  • Country: gb
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #13 on: October 08, 2015, 07:45:25 pm »
I did a couple of native 4k videos on Youtube a few months ago to see how it worked out.

In the end, the massive increase in transcoding time at my end meant that I really didn't consider it worthwhile doing longer term for now. In addition, I only have facilities for 25 fps at 4k, I don't know if Youtube can cope with 50 or 60 fps on 4k: 1080p at 50 or 60 fps seemed a reasonable enough compromise.
There are some 60fps 4k videos on youtube. I seem to remember we watched some which are 60fps and 3D at 4k, too.
 

Offline rs20

  • Super Contributor
  • ***
  • Posts: 2047
  • Country: au
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #14 on: October 08, 2015, 08:22:56 pm »
Is there a single low quality conversion in place that Dave is seeking to improve upon? Wouldn't Dave be better off just doing a single high quality conversion? If he is not doing that already.
The claim underlying this whole thread is that the re-encoding that YouTube does on their servers to (almost) all uploaded videos is this single "low quality" conversion in place.
 

Offline Spamlobster

  • Newbie
  • Posts: 4
  • Country: 00
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #15 on: October 08, 2015, 08:24:44 pm »
I'd think the upscaled solution might give better result only in still shots. The small increment in image quality & lesser artifacts - per frame - will likely never give nicer result if there's any real motion in the shot compared to 1080@50/60. And Dave, you're hyper, there's basicly always movement (camera or subject) in your shots.

#803 and #804 are a fresh example what the difference in fps alone does. (#804 was propably shot @50 fps but rendered @30, no?)

Is the small & situational image quality improvement worth the motion performance, absolutely not, not to mention the extra hassle of rendering etcetera.


Btw thanks for your work on EEVBlog.
 

Offline aerobaticant

  • Newbie
  • Posts: 2
  • Country: gb
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #16 on: October 08, 2015, 09:56:27 pm »
Hi Dave,

For what it's worth, if you shoot in 1080P then upload in 1080P. If the 1080P video is viewed full screen on a 2160P monitor then the monitor or video player will probably scale to 2160P uncompressed. This is almost certainly going to be better than a transcode.

If you were to shoot in 4K (2160P) then it would be worth uploading in 4K.

By the way I have been working in professional video for over 20 years.
 

Online coppice

  • Super Contributor
  • ***
  • Posts: 3675
  • Country: gb
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #17 on: October 09, 2015, 01:15:58 am »
Neither video looks that great, but there are definitely differences between them, at least when viewed on my 4k screen where the 4k version needs no further manipulation before it is displayed. The flaws in Dave's skin show up much better with the 4k version  :) The black shirt looks awful in both videos, although slightly less horrible in the 4k one. It quantises in a very unpleasant way.
 

Offline The Magic Rabbit

  • Regular Contributor
  • *
  • Posts: 108
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #18 on: October 09, 2015, 01:46:48 am »
The AVCHD compression is going to destroy any real difference. The colour depth and shadow detail is lacking, and the lighting doesn't help. Quite frankly it would be better to stop worrying about 4K when you're not able to fully exploit 1080P anyway.

I don't know what Coppice is talking about above. Both videos are 1080p via YouTube, so the idea only one needs manipulation is nonsense.

Sent from my XT1039 using Tapatalk

 

Online coppice

  • Super Contributor
  • ***
  • Posts: 3675
  • Country: gb
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #19 on: October 09, 2015, 01:49:11 am »
The AVCHD compression is going to destroy any real difference. The colour depth and shadow detail is lacking, and the lighting doesn't help. Quite frankly it would be better to stop worrying about 4K when you're not able to fully exploit 1080P anyway.

I don't know what Coppice is talking about above. Both videos are 1080p via YouTube, so the idea only one needs manipulation is nonsense.

Sent from my XT1039 using Tapatalk
I get the 4k video at 4k from youtube, and display it on a 4k monitor.
 

Offline The Magic Rabbit

  • Regular Contributor
  • *
  • Posts: 108
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #20 on: October 09, 2015, 01:55:15 am »
Interesting... I don't get a 4K option on either video.

Sent from my XT1039 using Tapatalk

 

Online coppice

  • Super Contributor
  • ***
  • Posts: 3675
  • Country: gb
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #21 on: October 09, 2015, 01:56:04 am »
Interesting... I don't get a 4K option on either video.

Sent from my XT1039 using Tapatalk
Firefox doesn't offer me a 4k option. Chrome does.
 

Offline The Magic Rabbit

  • Regular Contributor
  • *
  • Posts: 108
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #22 on: October 09, 2015, 01:57:01 am »
I'm using Chrome on Win7 64bit.

Sent from my XT1039 using Tapatalk

 

Online coppice

  • Super Contributor
  • ***
  • Posts: 3675
  • Country: gb
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #23 on: October 09, 2015, 01:59:01 am »
I'm using Chrome on Win7 64bit.
I'm using Fedora.
 

Offline The Magic Rabbit

  • Regular Contributor
  • *
  • Posts: 108
Re: eevBLAB #16 - 4K Render Youtube Experiment
« Reply #24 on: October 09, 2015, 02:02:09 am »
Well that's just showing off. Lol

I wonder if the You Tube player varies though... or maybe the content isn't at all resolutions in all countries.

I'm in the UK.

Sent from my XT1039 using Tapatalk

 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf