PDA

View Full Version : How to tell if a 3D point will be hidden



bigefrom84
12-21-2013, 04:08 AM
I want to place numbers next to vertices in a model only if that vertex is visible. I am using outline fonts and they will be hidden by the foreground images, but often part of the text gets clipped by adjacent objects. I have an image of this but don't see how to include it in this post. It seems i can only insert images based on URL's.


So i thought if i know that a particular vertex is not hidden, i could disable the depth test so the text would always be viable.

tonyo_au
12-21-2013, 04:33 PM
I can think of one way but it seems a bit clumsy.

1) render view
2) render vertices as points with an occlusion query (ARB-occlusion-query )
3) for those vertices not occluded render your text with depth disabled

Dark Photon
12-22-2013, 12:36 PM
As always, there are several ways you can do this. tonyo_au's given you one. You can fetch the occlusion query results back, and use in an "if" expression in your app code to decide whether to draw the text. A incrementally more efficient approach is to use those same occlusion queries in combination with conditional rendering (glBeginConditionalRender()...glEndConditionalRend er()) to move the "if" check down into the GL driver, and possibly onto the GPU.

And if you have a lot of vertices/texts to draw and you can preupload your text strings onto the GPU (in a texture, etc.), a potentially more efficient approach is to do your own occlusion checks on the GPU and generate your own list of "visible points" to render text for. Something like:

1) render view into FBO (depth buffer = texture)
2) Rebind the depth texture to an input sample of the pipeline
3) Render all of your vertices with transform feedback:
- Read depth buffer for vertex location.
- If vertex not occluded
- Stream out vertex ID to TF output buffer
4) Bind vertex output buffer as input to pipeline (as texture buffer, etc.)
5) Render your text strings for the generated vertex IDs

The key here is whether #5 can be pushed totally to the GPU with your text labels implementation. If not, ignore that last suggestion.