SceneCapture Camera performance - needs suggestions

Development Programming & Scripting
, , , , ,
1

Hey there! I’m facing the following:

If multiple SceneCapture cameras for external mirrors are used in a pawn the performance is dropping terribly. My question is how to make them more effective. What I tried/considering:

  • The tick interval is set to 0.1 instead of 0. In fact, any float above 0 is making the render much more speedy but the video obviously becames laggy. It needs improvement as images should fade to the next instead of overwiting?
  • The foliage always shows, no matter how far the MaxViewDistance is set. I don’t really need the foliage at all, skipable if it gains significant performance. How to exclude?
  • If ShowOnlyList is used, how can I set the white/blacklist for those cameras? Can’t find it in the docs either. By tag would be great.
  • Render those cameras in lower resolution? The renderTarget’s output resolution has no effect on speed.

As I don’t want to create an A/B problem, the situation is: I have several vehicle pawns having external mirrors. Most of the vehicles has to have at least 3-4 mirrors so each camera renders and the ProfileGPU shows each camera render is nearly takes as long as the main view, leaving fps around 20ish while 60+ is achived if no mirrors. Actual mirrorish materials using PlanarReflection has the same problem I guess. Any recommendations?

Thanks for your time and attention! :wink:

2

Hello, @gram-o-phone! How are you? I hope you’re doing great!

The way a SceneCapture “captures the scene” is almost the same as rendering it with a camera.
Imagine that you’re essentially going through the entire rendering process for each SceneCapturer—from iterating through relevant geometry to populate the different buffers in the rendering pipeline, all the way to applying post-processing effects.

That’s why it directly impacts performance.

Now, I know there are parameters that allow you to control these things and determine which processes should be considered in the SceneCapturer’s rendering flow.

I tested this a long time ago, and I’m pretty sure those settings are under “General Show Flags.” You can find more details in this video:

Youtube video link: https://www.youtube.com/watch?v=8cDjXxpQN9w

I know it’s not much, but I hope it helps in some way! :blush:

1 Like
3

Hey, you are everywhere. In the first place, I have to thank you helping by email to find out that the SceneCaptures are my biggest holdback in terms of fps. Only from Insights it was not obvious, in retrospective it is. :wink:

Also, I found the exact same video of yours and made several modification, so salute to you! :smile:

On the topic:

  • The biggest improvement and basically my final solution is to modify the mirror cameras tick time dinamically. I mapped the values between 0.05 to 0.4 tick time from 60 fps to 30fps (with target number smoothing to avoid big jumps), so it could be smooth enough if other circumstances are right. Of course, 0.4 seems ridiculously laggy, but hey, the main view should be perfect.
  • Offing the ‘render every frame’ in my case made things worse.
  • The size of renderTarget is not a huge factor (as you predicted), I found I can leave it to 2048
  • The show flags are important in my case, however lot of details has to remain.
  • Finally decided to get rid of Nanite. It was a small addon to our production value, returned to proper LODs for foliage. Combined with relative short render distance it made a huge difference.

With this combined from around 25 laggy fps it went to 40-50, or above in some cases. :+1: (4k, 4 local multiplayers for one vehicle as multiscreen, with 4 mirrors, i5/4070TiSuper)

(This looong text is for the future, probably others can benefit. By the amount and quality of your videos it’s a drop to the ocean of knowledge of yours. :wink: )