Andrew Pile from Vimeo.com Staff said the following the aforementioned thread:

“When I actually looked at the videos it all made sense: they haven’t upped the bitrate much, just the resolution. Block for block these videos are lower quality than 720p. We have tested 1080p video in the past and it played as well as YouTube during our tests (ie not well). This is a publicity stunt, less than 10% of people who even visit Vimeo have monitors capable of displaying 1080p resolution, where as over 80% are capable of 720p, and I’m sure its even worse for YouTube.”

My findings over the years has been the same.  Very few people who visit lakedance.com or other websites I currently run, have resolutions greater than 1600×1200, meaning they couldn’t see true 1080p quality even if they wanted to, and my theory is that Youtube is simply putting one over on people by using low bitrates and high resolution.

1080p streaming is a huge step forward for video distribution, no doubt about it.  Are they the first to do it?  No, but they’re Youtube and so when they adopt it, the world shakes.

So what’s so special about Youtube’s delivery of 1080p – well its optimized for streaming, as in it comes it at about 3700kbits/sec for streaming.  What that means is most people on a typical 3.5mbit or above (cable or DSL) connection should be able to watch with very little buffering or hiccups.  That’s great… but it MUST come at some cost… right?

I set out to see if my experience was right or whether Youtube had some how circumvented the seemingly unplausible streaming barrier – 1080p at great quality, and if so how did they do it?

Test: Download 2 different 1080p trailers from youtube.com and compare them to the same 1080p trailers downloaded from quicktime.com.  The Youtube trailers will be in a low bitrate mpeg-4, the quicktime files will be in a much higher bitrate .mov, both are 1080p.

Files Used:

1. Tron Legacy Concept Test – Youtube (71mb)  Quicktime (211mb)
2. The Dark Knight Trailer #3 – Youtube (64mb) Quicktime (197mb)

Programs Used:

1. VLC 1.0.3 for all file playback
2. Mozilla Firefox 3.6 beta with Download Helper to download Youtube files

Test System: AMD Phenom 9600 w/ 4GB Ram on 7mbit connection running Windows 7.

Additional Notes: Files were played back in VLC 1.0.3 but captures were done using Windows Screen Capture (ie printscreen).  Captures were loaded into Adobe Photoshop CS3, and saved at 73% jpeg, which appeared to have no visual loss from the screen capture.  These files have not been edited or altered in any way except to crop and save.

TRON LEGACY CONCEPT TEST

Tron Legacy by Disney is going to be a futuristic masterpiece.  It’s a standalone sequel to the 80’s original.  What’s interesting about the trailer is that it contains a LOT of large areas of similar color, ie. large expanses of black, white, or colors (such as the neons or lighting).  You only see this kind of situation in animation or heavily CGI enchanced films where there a lot of clean textures.   There’s not a lot of fine detail, and no film grain to speak of.  What this means is that it is a prime target for a highly compressed codec – so in this test, I expect the Youtube 1080p to hold its own, but lets find out.

Test 1 – Tron Wideshot – low motion – mixed detail

Tron Legacy Youtube 1080p 2:02

Tron Legacy - Apple QT 1080p - 2:02

In this shot you can see immediately that there is a loss in detail here.  While the Youtube frame holds up well in the larger expanses (white floor and ceiling), it fails to recognize many of the subtle details such as detail in the buildings in the distance, in the subject’s shirt, and along the walls.  Note that the little walls alongside the “ramp” have a design that is visible in the QT frame, this design is lost in the Youtube version. Additionally, note how much “crush” is happening in the shadows in the Youtube version, verses the details preserved in the blacks in the QT file.

Test 2 – Tron – high motion – high detail

Tron Legacy - Youtube 1080p - 1:52

Tron Legacy - Apple QT 1080p - 1:52

This shot takes place right after he crashes his bike, and the pieces you see below him are his exploded remnants.  There is a lot of detail here and while the Youtube frame doesn’t fail outright, again we can see the shortcuts it employs to accomplish its mission of low bitrate 1080p – there is significant loss in detail across the frame, and again the shadows have been crushed losing fine detail.  Most mid-range detail shows here but the obvious compression noise and jaggies along the leg of the rider in the Youtube frame show some serious faults.  Lets look at a side-by-side comparison.

Test 3 – Tron – Side-by-Side Comparison

Tron Legacy - Youtube 1080p vs Apple QT 1080p - 1:52

This should make it a little easier to compare what we were just discussing.  What’s interesting here is that entire shards of the broken bike are lost in the Youtube file, also look at the reflection on the guy’s back.  In the Youtube version it becomes so noisy we can’t even tell its a reflection.

THE DARK KNIGHT 1080p TRAILER #3 TEST

This trailer has almost every shot in motion, or a subject in motion.  It was hard to find much still frames to use for comparison.  Nevertheless I think I ended up with two very good examples.  Also, for certain shots I added a 720p version of the quicktime file for comparison.

Test 1 – The Dark Knight – Static Over-the-shoulder

The Dark Knight - Trailer 3 - Youtube 1080p - :54

The Dark Knight - Trailer 3 - Apple QT 1080p - :54

The Dark Knight - Trailer 3 - Apple QT 720p - :54

The Dark Knight - Trailer 3 - Side by Side

There’s a lot of opportunity here for the Youtube 1080p to succeed – large areas of similar colors – black mostly – but apparently that is not enough to keep our actresses face from looking airbrushed in comparison to the Apple QT files.  Note, even the Apple QT 720p file holds more detail on her face than the Youtube 1080p file does.  Lost luster can also been seen in the little reflection in the shiny strip around her neck.  But beyond that, again the crushing of detail in the shadows means we lose much of his hair details, and little other dress details.

The Dark Knight - Trailer 3 - Youtube 1080p - :59

The Dark Knight - Trailer 3 - Apple QT 1080p - :59

The Dark Knight - Trailer 3 - Youtube 720p - :59

The Dark Knight - Trailer 3 - Apple QT vs Youtube 1080p Side by Side - :59

Here’s where the Youtube 1080p encode absolutely fails (or at least gets a D- in comparison).  The actors faces are gone.  Totally indistinguishable.  Apple QT holds detail in both the 1080p and 720p versions.  Also note that the little buttons next to the “sunglasses” guy disappear in the Youtube 1080p file, the grating surrounding the heli pad loses fidelity, and the clothes of the people walking all lose substantial detail in Youtube’s encoding.

Well to be fair, on my 7mbit connection, the Youtube trailers seemed to stream almost flawlessly.  There was a hiccup but when I reloaded the file, it streamed fine.  Quicktime 1080p on the other hand, cannot stream.  However, where QT makes up for that is that you can download it in various sizes or straight to your ipod, whereas the Youtube file, unless you use a bypass downloader, is not going to ever be saved to your hard drive.  The two cross each other out, and, just for kicks, you can stream the lower sized versions of quicktime.  Neither source’s file chugged the computer, the Youtube .mp4 1080p file had a maximum 21% CPU and the QT 1080p file had 13% CPU.  Obviously there is something to this – more compression means higher CPU?

Conclusion

It’s one thing to pull stills from the video and tear them apart.  Its quite another to stream 1080p, regardless of the bitrate, straight into your computer.  In that regard, the Youtube 1080p experiment passes… it plays and it is, infact, 1080p.   But that’s not why I started this comparison.

In fact, I think we need to stop thinking about this as a comparison of Apple Quicktime vs.  Youtube 1080p.  This is a demonstration of how low of quality can we go, how close can we come to destroying an image that companies and artists spend millions to create, how much detail can we lose, before discerning viewers and companies take a stand.  In the last test, I demonstrated that the faces of the actors are removed in a high motion shot from Youtube.  In context, this happens in a fraction of a second, and indeed we can make out that there are people walking from a helipad somewhere else – but who are they?  The man closest to us whose face is absent in the Youtube version is Morgan Freeman.  This may not matter a great deal in a trailer but in a feature with so many characters and storylines to keep track of, it may be vital damning evidence of conspiracy or a vital clue of a killer in a mystery.  In the other examples – lost designs on walls, lost skin on a female lead, and lost shards of bike means you’re not getting the whole picture or the whole movie.

There’s no doubt, the Youtube 1080p files look much better than standard definition, but if you notice, it doesn’t even hold as much detail as the Quicktime 720p.  That is a big problem.  It’s not such a big problem for quickie views of new trailers, but for feature presentations it means the Youtube 1080p delivery just ain’t good enough yet.  And when the Vimeo guys say “It just doesn’t add up”, I believe them.  Until we can achieve a better picture quality than what we have seen from Youtube 1080p, we’re just not getting the whole movie the way it was meant to be seen.  Close, but no cigar.