checking in

I'm still working on LOD issues. At the moment I'm still puzzling over screen error estimation. I have code which works well most of the time, but breaks occasionally. There is no pattern to the failures, and circumstances which result in a poor estimate will work fine at other times. I wonder if I've stumbled onto a chaotic system at times. I'm even questioning if the ray casting into terrain is working properly (as I use it to test).

The testing methodology is to cast two rays from the camera into terrain, noting the screen positions of the castings. This gives two sets of screen positions with matching points in 3-space. Screen error estimation is performed on the two three-space points and compared with the delta between the actual screen positions.

I have a large collection of these sets saved off, and use them to batch test my algorithm. I then import all of the data into a statistics package for analysis.

At the moment I'm considering breaking out a software renderer I wrote years ago and using it to avoid the raycasting (since it obviously knows where three-space points end-up on the screen).

I'm not sure if Tuan ever received the files I sent on 12/1, hopefully those got through and he is just busy.

Here's hoping that everyone is enjoying the holidays.

Files received, must still take time to look at.

Unless "RSQ_Height_no_interpolation" rayscene query mask is set, casting a ray using PLSM2 scene manager will take plsm2 lod algo in account which may differs from your algorithm.

Did you get anywhere with those files?

Forgot about those, sorry !

Can you remind me the problem and where to look ?
(cam position would help, if it was about a "pop" problem, as I didn't found it on terrain. (nice one btw, which terrain modeler is that ?))

It has been a while, so I'll be a bit vague.

Turn and face the hill (to your left I believe) and start towards it. There will be severe terrain pops after a few seconds of travel.

The terrain was generated by a home-made generator I wrote in October. The generator isn't finished, since I wanted to be able to walk around the maps and critique it.

I've seen some popping, here's where and why :

When arriving near a tile border which is a highest lod level, and next tile is at lowest renderlevel, "stitching" make current tile border becoming "lower lod", hence having a huge pop, from highest to lowest...

That's indeed a problem, I don't see any easy solution to this. Have to think about it.

If it's not that sort of popping you'll have to give me cam corrdinates...