In my own subjective testing some years ago, it did not seem like increasing the resolution helped bring down the compression artefacts on youtube (tried with 2K on a 1080 monitor). But then again, times may have changed. But I don't think it's going to help much since youtube butchers bitrates for any content uploaded (although it's even worse for sub-720).
If there's an overwhelming demand for higher resolutions, then I'd say it might be worth it, but not right now since there's practically little gain for a lot longer compression time. This is just my own opinion, of course.
Not only resolutions above 1080p on youtube have noticeable less compression artifacts, but if you watch 4k video on 1080 monitor it looks noticeable sharper and cleaner than same video at 1080p. It's better to have those resolutions even for 2D games, in my opinion ofc, because there noticeable difference
edit:
http://i.imgur.com/Uz7DPyC.png
here it is. Both screenshots made at 1080 monitor with my lame encode. First made at 4k youtube replay and second second at 1080p replay. Both zoomed x8 with nearest neighbor. You can clearly see the difference. I can't see why not to have 4k encode if there is noticeable difference even at 1080 monitor. If you have 4k encode watcher can choose what he/she wants. Maybe someone wants more sharpness and clearness of the picture. Why not to have this?
That's mostly because of 4:2:0 upscale negating the information loss through adjacent pixel color redundancy (and higher resolutions being a better source to downscale that back to the monitor resolution). If you make two or four adjacent pixels share color to save bandwidth, just quadruple their amount, and each pixel will retain its original color. Simple and elegant.
YT does set a higher bitrate target for larger resolutions as well, but ever since the move to VP9 the benefit for low-res videogames seems to be marginal in that respect. When YouTube starts supporting native 10-bit 4:4:4 color space, upscaling to extreme resolutions (if at all) won't be as much of a necessity.
Warp wrote:
Edit: I think I understand now: It's my avatar, isn't it? It makes me look angry.
If you make two or four adjacent pixels share color to save bandwidth, just quadruple their amount, and each pixel will retain its original color. Simple and elegant.
I don't exactly understand that. This video was upscaled like this:
x16 Nearest neighbor -> 2880x2160 bilinear interpolation
by using VirtualDub filters, because i'm to stupid. I can't understand that chroma samples and other stuff. Is it something wrong with this? By comparing with tasvideos 1080p encode - i can't see the difference. My encode at 1080p have 3 blended vertical color lines and tasvideos 1080p encode have the same. 4k have only 2 vertical blended color lines.
http://i.imgur.com/NxUh2Jy.png
https://en.wikipedia.org/wiki/Chrominance
The picture is especially enlightening.
Imagine a block of 4 pixels, or a 2x2 block of pixels. To save space, we remove 2 pixels of chrominance and 2 pixels of luminance. The resulting scheme is called 4:2:0. On decompression, we will try to "guess" the color information we threw away.
But if you double the pixels, then we'll actually not throw away any information, because the pixels we throw away will be duplicates of the real pixels, so we lose no information.
That's basically the gist of it, though I may have gotten some details wrong.
EDIT: The pictures here (https://en.wikipedia.org/wiki/Chroma_subsampling) under Sampling systems and ratios is also enlightening.
Joined: 4/17/2010
Posts: 11475
Location: Lake Chargoggagoggmanchauggagoggchaubunagungamaugg
Koh1fds wrote:
Not only resolutions above 1080p on youtube have noticeable less compression artifacts, but if you watch 4k video on 1080 monitor it looks noticeable sharper and cleaner than same video at 1080p. It's better to have those resolutions even for 2D games, in my opinion ofc, because there noticeable difference
edit:
http://i.imgur.com/Uz7DPyC.png
here it is. Both screenshots made at 1080 monitor with my lame encode. First made at 4k youtube replay and second second at 1080p replay. Both zoomed x8 with nearest neighbor. You can clearly see the difference. I can't see why not to have 4k encode if there is noticeable difference even at 1080 monitor. If you have 4k encode watcher can choose what he/she wants. Maybe someone wants more sharpness and clearness of the picture. Why not to have this?
This is not about whether there are artifacts or not, because there mostly are anyway. This is about how much does your naked eye spot compared to lossless encode. You'll never get the true lossless look out of youtube, no matter how much "better" it gets. Transparency is when it stops making the difference for an eye.
We were uploading 2D encodes that got processed up to 4K, even though their actual res was lower: 224*8 for instance. Their 4K res is artificial if you want pixel-perfection, the only thing it guarantees is higher bitrate, which is still lossy, so I dunno.
If someone finally does mkvmerge tests and compares mkv aspect ratio correction against avisynth one, and the former doesn't look worse, we will switch to that, since then we won't generate too much overhead going 8x again. Otherwise... it's not worth it.
PS: Your comparison is pointless, you should have put 1440p back to back with 4K, since we do upload 1440p+ minimally. I set up the official script to do that.
Warning: When making decisions, I try to collect as much data as possible before actually deciding. I try to abstract away and see the principles behind real world events and people's opinions. I try to generalize them and turn into something clear and reusable. I hate depending on unpredictable and having to make lottery guesses. Any problem can be solved by systems thinking and acting.
PS: Your comparison is pointless, you should have put 1440p back to back with 4K, since we do upload 1440p+ minimally. I set up the official script to do that.
I didn't know that. Super C TAS encode have 1080p max. At least with HTML5 youtube version.
edit: here it is. Youtube finally proceed stupid video.
https://www.youtube.com/watch?v=Rpo_CloqZ2c
On 1080p monitor 4k is much better. Even this is extreme condition. I still thinks that 4k is good to have. Even there is not that big difference between 4k and 1440p than between 1440p and 1080p. 4k still the most cleaner of all on 1080p monitor.
edit2: and i don't say that you MUST right now encode everything at 4k right now. I just say that there is a noticeable difference. So it's not completly worthless
Joined: 4/17/2010
Posts: 11475
Location: Lake Chargoggagoggmanchauggagoggchaubunagungamaugg
For whatever schizophrenic reason they just drop the res by 1 when you're uploading 60fps. For html5. We upload 1440p+ anyway.
Koh1fds wrote:
On 1080p monitor 4k is much better. Even this is extreme condition. I still thinks that 4k is good to have. Even there is not that big difference between 4k and 1440p than between 1440p and 1080p. 4k still the most cleaner of all on 1080p monitor
I posted the deal above.
Warning: When making decisions, I try to collect as much data as possible before actually deciding. I try to abstract away and see the principles behind real world events and people's opinions. I try to generalize them and turn into something clear and reusable. I hate depending on unpredictable and having to make lottery guesses. Any problem can be solved by systems thinking and acting.
For whatever schizophrenic reason they just drop the res by 1 when you're uploading 60fps. For html5. We upload 1440p+ anyway.
I don't know how this exactly works. I uploads before x16 videos and i get only 1080p at max with 60 fps html5 player. If i uploads exactly 2160 height video i get 4k 60fps. Dunno why
Joined: 4/17/2010
Posts: 11475
Location: Lake Chargoggagoggmanchauggagoggchaubunagungamaugg
You can not understand a schizophrenic, that's why.
Warning: When making decisions, I try to collect as much data as possible before actually deciding. I try to abstract away and see the principles behind real world events and people's opinions. I try to generalize them and turn into something clear and reusable. I hate depending on unpredictable and having to make lottery guesses. Any problem can be solved by systems thinking and acting.
And this is how it works for 1.5 years now. So i don't think that those tases that has only 1080p60fps would be re-encoded by youtube to have 1440p60fps any time soon. And obviously everyone wants that 60fps. So basically at least for now my comparison was correct.
Joined: 4/17/2010
Posts: 11475
Location: Lake Chargoggagoggmanchauggagoggchaubunagungamaugg
True. But targeting at specific res will make the file size just as huge. Need to see how different our current method will be compared to aiming at 1440p exactly. Can you make a 1440p@60 test upload just to make sure it works that way for any res, not just 4K?
Warning: When making decisions, I try to collect as much data as possible before actually deciding. I try to abstract away and see the principles behind real world events and people's opinions. I try to generalize them and turn into something clear and reusable. I hate depending on unpredictable and having to make lottery guesses. Any problem can be solved by systems thinking and acting.
Joined: 4/17/2010
Posts: 11475
Location: Lake Chargoggagoggmanchauggagoggchaubunagungamaugg
Not sure what you mean, in virtualdub just set the target height and it will calculate the rest for you.
Warning: When making decisions, I try to collect as much data as possible before actually deciding. I try to abstract away and see the principles behind real world events and people's opinions. I try to generalize them and turn into something clear and reusable. I hate depending on unpredictable and having to make lottery guesses. Any problem can be solved by systems thinking and acting.
Why are you downscaling to 1920x1440 instead of just passing 2048x1792 to your encoder? Once you leave nearest neighbor 2x upscaling you're opening yourself up to loss since you no longer have the buffer zone for the yuv420 downsample to eat into.
Why are you downscaling to 1920x1440 instead of just passing 2048x1792 to your encoder? Once you leave nearest neighbor 2x upscaling you're opening yourself up to loss since you no longer have the buffer zone for the yuv420 downsample to eat into.
Hmmm... Because if you upload on youtube 2048x1792 video you get only 1080p60fps at max. If you upload video EXACTLY with 1440 pixels height you get 1440p60fps at max. Same with 4k. You have to upload EXACTLY 2160 height video. (or maybe close to it). If i do 2048x1792 to 1920x1440 downscaling incorrectly that's just because i don't know how to do it correctly
Joined: 4/17/2010
Posts: 11475
Location: Lake Chargoggagoggmanchauggagoggchaubunagungamaugg
Both vids are still 1080p max.
Warning: When making decisions, I try to collect as much data as possible before actually deciding. I try to abstract away and see the principles behind real world events and people's opinions. I try to generalize them and turn into something clear and reusable. I hate depending on unpredictable and having to make lottery guesses. Any problem can be solved by systems thinking and acting.
Joined: 4/17/2010
Posts: 11475
Location: Lake Chargoggagoggmanchauggagoggchaubunagungamaugg
But not for me... Anyone else?
Warning: When making decisions, I try to collect as much data as possible before actually deciding. I try to abstract away and see the principles behind real world events and people's opinions. I try to generalize them and turn into something clear and reusable. I hate depending on unpredictable and having to make lottery guesses. Any problem can be solved by systems thinking and acting.
Joined: 4/17/2010
Posts: 11475
Location: Lake Chargoggagoggmanchauggagoggchaubunagungamaugg
In firefox I also see 1440p. Alright.
Warning: When making decisions, I try to collect as much data as possible before actually deciding. I try to abstract away and see the principles behind real world events and people's opinions. I try to generalize them and turn into something clear and reusable. I hate depending on unpredictable and having to make lottery guesses. Any problem can be solved by systems thinking and acting.
So maybe youtube made 3 sets of encodes. Like at first it made sets of encodes for flash player that have only 30 fps encodes. Than it made another set that have 1080p60fps at max for html5 player. Than it made third set of encodes for html5 that have like 1440p60fps and 4k60 fps. Maybe third set uses different encode settings or even different codec.
Joined: 4/17/2010
Posts: 11475
Location: Lake Chargoggagoggmanchauggagoggchaubunagungamaugg
Yeah, whatever, it now forces everyone to use htmp5, and in newest browsers it supposedly shows 1440p.
Warning: When making decisions, I try to collect as much data as possible before actually deciding. I try to abstract away and see the principles behind real world events and people's opinions. I try to generalize them and turn into something clear and reusable. I hate depending on unpredictable and having to make lottery guesses. Any problem can be solved by systems thinking and acting.
