Development notes

Update on 360 capture for VRCHIVE

This week I’ve been working on a small contract for 360-degree panorama hosting website VRCHIVE to create a free Unity plugin that enables the player to capture and upload 360-degree panorama screenshots. You can check out a sample here:

http://alpha.vrchive.com/image/rC

In principle it’s pretty straightforward – I use RenderToCubemap to create a cubemap, then transform it into the equirectangular transform required by VRCHIVE, then upload it using the Chevereto API.

This approach works just fine for small panos, but there are a number of scalability issues as I move to larger panos: the time required for the equirectangular transform grows linearly with the number of pixels, and even an optimized CPU version can take on the order of minutes. I can make it run asynchronously with the game using coroutines, but this slows it down even further.

I created a version of the transform that runs on GPU using compute shaders, which is about 30x faster and can capture a 8192×4096 pano in about 2 seconds on my GTX 980, but these aren’t universally available and I’m encountering some issues where it fails for people on other GPUs (weirdly it works for me both on my fancy GTX 980 desktop PC and my 5-year-old laptop – I may have to try switching to integrated graphics for testing). I believe texture memory consumption may be at fault, as I can reproduce similar failures when Unity’s RAM usage is very high – so I’m investigating various ways to capture the pano using less storage. Unfortunately these too result in processing slowdown.

Transforming the panos to equirectangular on server-side would be more reliable and faster for the player, but also is relatively expensive for a small startup site like VRCHIVE. Another plausible alternative is direct support for cubemap panos in the VRCHIVE viewer.

I haven’t yet addressed the complications raised by e.g. the use of multiple camera rigs with different layer settings, which are prevalent in several VR titles that I know of. I can’t directly call RenderToCubemap on those cameras because they have the wrong settings for it – so I may have to create a separate cubemap camera and clone their layer settings.

Current focus is fixing the failures on other GPUs – luckily my eVRydayVR fan chat has lots of great testers, so I think I can find ways to fix this!

Meanwhile, SVVR is starting tomorrow and I have to prepare for that too. And I just got back this week from the Vive Jam where I got to help with Unity scripting on Electro Smash, a collaboration with several others there (mostly from Virtuix). So many exciting things going on!

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s