Best size of source for webvr

Q&A about the latest versions
Post Reply
JSchrader
Posts: 141
Joined: Sat Nov 28, 2015 2:26 pm

Pano2VR 6, Beta 2:
When comparing a 6k pano sideloaded to the Oculus Go with the same 6k pano processed through Pano2VR WebVR and viewed with Oculus Browser, the first one is visibly sharper.
When I use a 20k version from the same scene in Pano2VR instead, the result is even more blurry and useless.

Is there an input size that you guys optimized your viewer for?

What are the experiences from others?
User avatar
thomas
Chief Gnome
Posts: 2611
Joined: Fri Sep 01, 2006 3:56 pm
Location: Vienna, Austria
Contact:

Currently, we are still testing different ways to improve image quality and resolution, but as stereo is still missing, it doesn't make sense to start the final tweaking of the rendering.
One way to produce sharper images (but also risk aliasing) is to change the interpolation in the Settings/Preferences. Lanczos2 is much sharper, but as aliasing in a VR headset is much worse than on a normal screen (at least for me) as the flickering is uneven across the eyes. Lanczos3 should be a good compromise, but this all comes down to personal taste.
Screenshot 2019-04-02 11.20.04.png
Screenshot 2019-04-02 11.20.04.png (27.28 KiB) Viewed 1600 times
Another huge problem is the hardware limitations on the Oculus Go, especially with WebVR. As they wanted a cheap device they went with a Snapdragon 821 processor from 2016 so there is not a lot of headroom to do some additional things with the image. With WebVR you can not directly access the physical screen image and everything runs thru the ATW (Asynchronous Time Warp) that produces an addition remapping step. If you test the same output with a Lenovo Mirage Solo and Chrome (Dev build) the image is much sharper and there is a lot of headroom in the rendering as it has a faster SoC and I guess Chrome has better optimizations. I also guess the ATW adds some sharpening. For everything else, the Mirage Solo is not a very good experience and running a Dev version as Browser is not a good idea. I also like the screen on the Samsung Odyssey+ as it has almost no screen door effect. As there the pixels touch each other, aliasing looks even worse, as you get blocky zigzag lines. Finding a good balance for different HMDs is hard.
At the moment the rendering is optimized for multiresolution, and the player selects the proper level depending on the device, FoV, render buffer size,... A switch bias is currently also missing, as are other things, that are planned for the final release. The first beta was planned to get feedback on the user interaction (rotation with touch etc. ), but the next betas will focus on image quality and rendering, as this tweaking is way more time consuming, and the time and effort would have been wasted if we got the user interaction wrong. Since beta 2 also the single resolution output is supported. There we don't apply mipmapping, so you can experiment with different cube face sizes. The problem here as this is only good for one device, and the findings may not apply to other headsets.

TL;DR: We are on it. Please give us some time. It is still a beta.
MfG, Thomas
JSchrader
Posts: 141
Joined: Sat Nov 28, 2015 2:26 pm

Thank you Thomas, for the detailed reply.
I highly appreciate your dedication, and I understand that it's not possible to have all the good things at once :mrgreen:
I continue testing and will report back.
Post Reply