CV VRcam presets vs Youtube help info (and 360vid issues in general)
Posted: 10 January 2017 10:20 AM   [ Ignore ]  
Total Posts:  2
Joined  2013-09-20

Just starting to dig in to 360 vids now with CV-VRcam and CV tutorials and one thing I’ve noticed is a discrepancy I’m seeing between what Youtube says videos should be, and what CV tutorials and CV-VRcam think. Youtube wants 1:1 videos for VR and 2:1 for 360, but CVtuts and VRcam talk about everything 16:9. It works to use CV’s approach, but I’m wondering if there is a better “sweet spot” for appropriate and acceptable resolution and manageable render times and file sizes. I’ve done a bunch of tests and sometimes get worse results (once uploaded to Youtube) from higher resolution, and the VRcam presets only give OK quality.

I’m new to most of this so I might just be missing something, so I thought I’d throw it our here and see if people have any insight.

Youtube help for 360 videos says:

“We recommend uploading 360 videos (equirectangular format with a 2:1 aspect ratio) at a resolution of 7168x3584 or higher, up to 8192x4096.”

(from: https://support.google.com/youtube/answer/6178631?hl=en&ref_topic=2888648)

Youtube help for VR videos says:

“Before you upload, we recommend that you export your content in over-under equirectangular format with a 1:1 aspect ratio at a resolution of 5120x5120 or higher, up to 8192x8192. Maintain square pixels (i.e.,1:1 pixel/scale aspect ratio).”

(from: https://support.google.com/youtube/answer/6316263?hl=en&ref_topic=2888648

Profile
 
 
Posted: 10 January 2017 11:54 AM   [ Ignore ]   [ # 1 ]  
Moderator
Avatar
Total Posts:  4271
Joined  2011-03-03

Hi mikehoium,

The compression of any footage is very much based on its visual content. Per image and over time. The compression method one choses is either a good fit or works against the material.

There key is to get the most data to YouTube but also in a way that they do not “recompress” already heavily compress data. To upload uncompressed seems often, especially with 8K x8K prohibitive.

The best way to get good quality is often to increase the frame rate, which is also the base for a better viewing experience. Since in both cases, if a head rotation will result in different images, motion blur is out of the window.
But in normal cinema footage that is part of any fast moving content (camera based on object based). Motion blur “eats” details and requires so less data.

Often a stream is not based on content, it is just limited by a fixed max bitrate.

In these few lines I have reflected my personal understanding of it. In a nut shell, the problems of getting a great quality is based on how much data can be used per second, what frame rate is used, how large is the image and how much change and details are in the content.


If you go now and do your tests, use something that comes close to your wanted content, and use always the same scene. After uploading, wait a while, that the movie streams already (so my experience) is not the moment where YouTube is done with it. You can see that often, that HD is available but 4K (UHD) is not.

To have 16:9 or 2:1 as an option is certainly a recommendation of the material one has natively. If the camera set up (practically) gives you an equirectangular 2:1, one would lower the quality by changing it to 16:9, vice versa.
It is like with any change, e.g., lens distortion correction: any pixel move less than a complete and precise pixel-distance will result in less information (as in contrast, sharpness, etc). Since that blurs to a certain degree the content, the compression has less work to do. IF one sharpens then in post the content to get a visual quality (kind of) back, the pronounced edges, as that what you get, is not real content sharpness but visually forced sharpness. Such harm to the content will be paid in more need for bandwidth, since it is often already maxed out, the quality goes even more down. If the content is then oversampled, even smaller resolutions(here the pixel amount) can result in more bandwidth need—considering Mbit/pixel ratio.

Here I see a reason why YouTube provides the option to give you both ratio options, and doesn’t force you to convert and lower the quality.

Since we talk about streaming, the end user device and encoder on the other side is in the game as well, it might vary drastically.

All of that is a formula where I have no general answer to, and even the presets from Adobe haven’t given me pleasure for all footage, as these presets have no idea about “my” content. So it is a little bit trial and error.

Set up higher frame rates (yes, I know more render time), or try to find your compression settings where you can’t see any loss, go a little bit higher perhaps for delivery. What your videos (if possible) on a scree 1:1. If you have only 2K or something, watch in After Effects in 100%. Often even 5K monitors are set to 2.5K to keep software interfaces readable, and QuickTime can’t show then 4K content, it needs to be switched to 1:1 resolution.

I hope I could point you to some areas where to look, and what matters. Since highly detailed compress differently than low detail (per image or over time), make your tests. Keep in mind that one often can see up to (or even above) 90º, which is a quarter of the footage provided. IF you use a 2K phone as viewer, anything below 4K will fail, but anything *K might kill the stream. Tests, tests, test.

Something similar was discussed here:
https://www.cineversity.com/forums/viewthread/1890/

All the best


P.S.: I checked [again] the availability of books for that—which I do frequently, but the latest about compression I could find was from 2010: nothing new. Assuming the research was held in 2009 and before for this book, the content is stone-age in a web-based age.
The articles found while google the theme, are mostly a year old and the majority of them tried to sell software solutions, i.e., presumable biased information. The first book I bought about compression options and techniques was in the ‘90s, and even that book has seen no updates. In other words, the development in this area is obviously too fast to make anything in print available.

 Signature 

Dr. Sassi V. Sassmannshausen Ph.D.

Photography For C4D Artists: 200 Free Tutorials: Texture, Panorama, HDRI, Camera Projection, etc.
https://www.youtube.com/user/DrSassiLA/playlists

Profile
 
 
Posted: 11 January 2017 06:46 AM   [ Ignore ]   [ # 2 ]  
Administrator
Avatar
Total Posts:  321
Joined  2006-01-24

Hi Mike,

In general, it’s best for monoscopic 360 images to have a 2:1 ratio (4096x2048) and stereoscopic 360 VR to have a 1:1 ratio (4096x4096). You have to adjust for different output though in order to use as much resolution as you can. Most phones can’t play back a 4096x4096 video for instance, so the compromise is often to do everything 4096x2048 or UHD.

YouTube is based around a 16:9 ratio, and automatically recompresses everything you upload into multiple 16:9 sizes (UHD, 1080p, 720p, 480p). If you upload a 2:1 image, it scale-fits it into that 16:9 box, causing a letterbox. The letterbox isn’t visible in VR mode, but it’s nonetheless a waste of pixels in a situation where we need as much resolution as we can get. As a result, I recommend uploading YouTube 360 and VR content at UHD resolution. I think the YouTube white paper is referring to the ideal situation from a general academic standpoint, without accounting for their own output environment.

I do think the CV-VRCam presets should be tweaked, because in general it’s best to render at the highest possible resolution and a 2:1 aspect ratio, and squeeze the video for delivery after compositing. But of course you have to balance it all with the render time required to output video at such high resolution.

Profile
 
 
Posted: 11 January 2017 08:38 AM   [ Ignore ]   [ # 3 ]  
Total Posts:  2
Joined  2013-09-20

Thanks Dr Sassi and Rick

That is the specific sort of info I’m looking for. It is weird that Youtube’s own help files don’t account for how they actually do things in their system.

Seems as 360 vids might be a big thing - great if CV-VRcam got tweaked, and maybe more tutorials on workflow and various issues.

Mike

Profile
 
 
Posted: 11 January 2017 09:19 AM   [ Ignore ]   [ # 4 ]  
Moderator
Avatar
Total Posts:  4271
Joined  2011-03-03

Hi Mike,

You’re welcome, and having been for a while on the Google+ Create Group, I got the feeling it is a constant changing direction over there. (Google owns YouTube).

Having said that, the VR tech-specs are certainly not fixed like a “one size fits all” standard, and it might take a while before the changes slow down, IMHO, but please: post in the Tutorial Suggestion forum what you would like to see.

As usual you can’t go wrong with Rick’s information. Try to give YouTube no (or very little) option to change your material, change is loss.

My best wishes

 Signature 

Dr. Sassi V. Sassmannshausen Ph.D.

Photography For C4D Artists: 200 Free Tutorials: Texture, Panorama, HDRI, Camera Projection, etc.
https://www.youtube.com/user/DrSassiLA/playlists

Profile