• Advertisement
Sign in to follow this  

Deferred rendering basics

This topic is 1920 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

Hello,
I'm trying to implement deferred rendering and I'm stuck with this problem. I do all calculations in view space, I set the light position for testing to (0,0,0) - camera position. But the problem is when I look up or down the lighting changes to brighter or darker. It sounds like I have something wrong with matrices but I checked that many times and still no luck. I'm not sure if my reconstruction from depth is right because it's a new thing to me and I may be doing something wrong here. But on the other hand lighting doesn't change when I move backward or forward and keep the angle (so I guess position is correct?). Anyway I'm posting some pictures, maybe someone can figure out where I go wrong. For now I'm trying to compute lighting with geometry normals and then move with bump maps. To me my normals look too bright, pinkish, dull etc. when I compare them to normals from Killzone paper or other tutorials. Maybe this is the case?

Thanks for watching

Edit: I consider simplest scenario for now - only diffuse lighting, no attenuation, no specular and no ambient

This is how I recreate position from depth in light pass:
[CODE]
float z = f1tex2D(depth_map, texCoord);
float x = texCoord.x * 2 - 1;
float y = (1 - texCoord.y) * 2 - 1;
float4 vProjectedPos = float4(x, y, z, 1.0f);
//unproject the position from clip space to view space using inv. proj. matrix
float4 vPositionVS = mul(invProj, vProjectedPos);
float3 vsPos = vPositionVS.xyz / vPositionVS.w;
[/CODE]

I use depth buffer to get depth or I can use depth encoded as color which is:
in vertex program:
[CODE]outPosition = mul(ModelViewProj, position);
vDepthCS.xy = outPosition.zw;[/CODE]

and in fragment program I do division z/w and pack it as color but the result is the same as with depth buffer. Edited by keym

Share this post


Link to post
Share on other sites
Advertisement
First off, you can't display the normals without hack, because normals have negative values. Therefore you need to map them to a visible colorspace (color = normal*0.5+0.5), I guess that killzone use an other mapping(or none at all).

Your normal screenshot seems to be ok (green pointing up, red pointing right, blue tint for pointing to the camera), how does the normals change if you look up/down, maybe your inversion of the y-coord is not consistent. Edited by Ashaman73

Share this post


Link to post
Share on other sites
Hi,
I solved my problem yesterday. Normals were ok (I packed and unpacked them from color with that hack you mention but I didn't mention that, that Killzone paper made me doubt...). Turned out that my reconstruction of position was wrong (like said before, it's a new thing to me and I was unsure of that bit the most). I used this [url="http://www.opengl.org/discussion_boards/showthread.php/177001-Easiest-way-to-reconstruct-eye-space-position?p=1234440&viewfull=1#post1234440"]http://www.opengl.or...l=1#post1234440[/url] code and it finally worked [img]http://public.gamedev.net//public/style_emoticons/default/smile.png[/img]. Thanks for interest though.

Cheers

Share this post


Link to post
Share on other sites
Sign in to follow this  

  • Advertisement