Sign in to follow this  
xwskx

Any website like, screen-space error metric for dumbies?

Recommended Posts

xwskx    122
Hello i'm a noob, I've been trying to develope and LOD algorithm, and intially I have been determining level by taking the distance from the camera to the terrain subset. But as i began researching the topic. I realized that screen-space error metric, is more ideal for determing the LOD. But for some odd reason, the concept doesn't click. I think the concept is something like taking the distance between the reference height and the vertex height to find the significance of the vertex. If its within the threshold than add the vertex, if not remove. Is this correct? Does anyone know any sites i can visit to learn this in detail. I've been doing searches, but they end up being only a small part of the document and many of them assumes you already know it. Thanks for the help.

Share this post


Link to post
Share on other sites
mrbastard    1577
I can't think of a site with a simple explanation, but I'll try and provide one.

Quote:
Original post by xwskx
taking the distance between the reference height and the vertex height to find the significance of the vertex. If its within the threshold than add the vertex, if not remove.

What you describe there is known as a geometric error metric - the distance is measured in 3d space.

A screen space error metric would project the two points (with the same matrix as the viewing transformation) onto the view plane and measure the distance in screen space. It's an oversimplification, but it can help to think of it as the difference in screen pixels - the number of pixels that will 'pop'.

It sounds like you're thinking of using a screen space error metric to decide which edges to collapse or which vertices to remove. This may be a bad idea, depending on your app; screenspace error is only really valid for the given viewing transformation. So if you were thinking of doing the LOD model creation in a preprocess instead of per frame (you were, weren't you? [smile]) you'd have to stick with the same position, orientation, FoV, near and far planes, etc.

There are some assumptions you can make, essentially trading accuracy for a simplified scale factor that's valid in most positions and orientations, instead of doing the actual camera transform which is only valid for that viewpoint.

It's more normal to record the geometric error (or an average over an area) and then project that to get your screenspace error. So you'd do your LOD model creation based on geometric error, choose which LOD model to use at runtime based on the screenspace error for the current view.

It's also worth bearing in mind that a screen space error metric is only really useful if your LOD models are good enough. With Geomipmaps, for instance, it's nearly pointless unless your terrain is smooth rolling hills. Any high frequency detail and your geometric (and thus screenspace) error is huge, because the vertex selection method for the LOD model is totally arbitrary.

You are right though - screen space error is one of the best indicators of the difference the user will perceive. It's just that for it to be practical there are some big things to consider.

I learnt about general LOD concepts from the Geomipmaps and ChunkedLOD papers, and then found more info than I will probably ever use in this book. It goes over the basics in fairly easy to understand terms and goes into quite a lot of detail on more complex things, but can seem a bit too abstract at first. It's not a 'fun' book by any stretch.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

Sign in to follow this