[Beta 3: Bugs] Still have shadow noise, some other items not working

#1 Turn down the shadow smoothing (avoid getting past 0.2). It starts artifacting when the sampling radius is too big. Also, try messing with the logWeight and shadowDistance to get some extra shadow resolution up-close, it looks very low.

And you'll always have a little noise: that's how the shadow smoothing works.

#2

Quote:the pixelization that occurs on the material

Manoel's post is spot on.

Also we do plan to implement a simple PCF filter before the final release. This will improve performance and provide a smooth shadow edge, but you will see more the the shadow resolution.

Some people seem to prefer that look to the noise... so you'll be able to do either.

Quote:2. rotation on a spawnPoint

I'm pretty sure it works... its just hard to see a sphere rotate when there isn't a point of reference.

There is a toggle on the toolbar that switches between rotating in world space or in object space... that has the side effect of not showing you the right rotation axis.

Quote:3. Unable to delete/remove Terrain Painter layers

I'll do that today... its been bugging me too and its really needed. I'll add duplication and rearrangement while i'm at it and make sure it can all undo/redo.

Quote:4. Occasionally I get a shadow cast on the terrain like there

I'm pretty sure i know what this is... it only happens when entering/exiting the editor. Its the player skinmesh not being updated and rendering garbage verts into the shadowmap blocking out the sunlight. When you turn around and see the player the skinmesh is updated and things work again as normal.

It's on my bug list.

#3 I was playing with the shadow shader, and enabled the non-diffused render path, since the diffusion effect is brutal on older hardware (testing on a G42-based Quadro).

I also did some tests with different target formats for the shadows (using R16F instead of R32F), and the performance improved greatly on this card, but there were some strange shadow artifacts that only went away if I tuned down the overdark, which made them too faint. Anyway, since this card can do filtering with R16F textures I noticed the shadows were filtered without PCF or something like that. Exponential shadow maps, heh?

#4

Quote:Exponential shadow maps, heh?

I'm really not using the ESM to avoid PCF... to do that i need to store two channels of depth (depth,depth^2) and do a more complicated calculation.

We use the ESM to cure shadow acne in a better way than a traditional bias value.

Quote:I also did some tests with different target formats for the shadows

R16F is gonna be too little precision for the PSSM shadow... we tried using that for a long time and the quality was horrible on even smaller scenes. Its a lost cause there i think.

But for normal sized spot and point lights a R16F will give you acceptable results with little image difference.

I'd love to get some hard performance data across a few different cards between R16F and R32F in undercity where we have like 40-50 shadowed point and spot lights.

#5

Quote:
I'm pretty sure it works... its just hard to see a sphere rotate when there isn't a point of reference.

I was just complaining to Steven about just this thing yesterday =)

#6

Quote:R16F is gonna be too little precision for the PSSM shadow... we tried using that for a long time and the quality was horrible on even smaller scenes. Its a lost cause there i think.

But for normal sized spot and point lights a R16F will give you acceptable results with little image difference.

I'd love to get some hard performance data across a few different cards between R16F and R32F in undercity where we have like 40-50 shadowed point and spot lights.

I don't have a Geforce7 to test, but on GF6 cards R32F is completely unusable. I get 15fps on even the most simple scenes, at 800x600, even with the shadow shader modified to not use the softening. The framerate is tied directly to how much lit pixels I have on screen: moving the camera so I see mostly the sky increases framerate into 60+ (as does disabling the shadows). When I switched to R16F, my framerate bumps to 45~50 fps. The GF6 can't even filter R32F textures.

Note that using the encoded R8G8B8A8 format was as fast as using R16F, but the encoding/decoding code didn't seem complete yet (couldn't get it to work properly), so that's definitely something to look into. I didn't test R16G16F, but I assume it performs just as well.

Its a bit hard to do some extensive tests on this since at home I got a HD 4850 that eats whatever the shadows use for breakfast, and at work I'm not dealing with graphics yet, so I'm limited to quick tests during breaks.

#7 I'll mess with the shadows. Thanks.

Yep, the rotation does work. Must not have really seen it rotating like you said.

3 NEW BUGS:

1.Once I change where my object drops. Center Screen to whatever... The object will drop in a crazy position 1000,1000,1000 away from my current location or back to 0,0,0. I was unable to change this back to Center Screen or any other option. It just kept throwing things way out there.

2.I'm getting a runtime error...crash on opening with the world editor button. It said it had a bad application call. Then named the executable. Doesn't happen all the time, only about every three times I run T3D, then it will crash a few times. Restarting doesn't help.

I'm not for sure if it was coincidence, but it happened after I put a function in using getTransform/change the distance/setTransform for a group of objects in a simset. I'll have to dig more...but someone might have an idea as to why this is happening.

3.I also noticed I had two copies of my project running at the same time in the task manager, even though I has not running T3D, just the toolbox. Maybe causing the problem above.

#8 @Manoel - Very good info.

I think i'll see about having some shadow quality settings which let you switch between R16F and R32F which should allow us to do better testing.

The RGBA8 format would only be a win if the performance is significantly better than R16F. Or maybe if R16F isn't filterable on a significant amount of cards.

So much to do. :)

#9 Edit on the 1st new bug I found...the offset problem when you add a new object to the scene...it seems to be on scripted objects only.