EternalNewbie

Member
  • Content count

    31
  • Joined

  • Last visited

Community Reputation

207 Neutral

About EternalNewbie

  • Rank
    Member
  1. SSAO artifacts

    Thanks tried changing the kernel size but this doesn't seem to make any difference. Adding a minimum depth difference helps to remove the artifacts in some areas but as I move the camera it simply reappears at different depths.
  2. SSAO artifacts

    Hi all I'm trying to implement very basic ssao but get the following artifacts. I'm not using the normals and I'm using a simple filter with no noise here. I've tried using a noise filter which helps remove them on the floor except when looking directly at them. However they seem to persist on the walls and other objects. Has anyone seen these sort of artifacts before? Any Idea how to get rid them? [attachment=11715:SSAOArtefacts.jpg]
  3. Get the current display mode using D3D10

    Quote:Original post by wavetable however, i tried D3D10CreateDeviceAndSwapChain() and it *only* works with a valid DXGI_SWAP_CHAIN_DESC here. does your code work without it? if it works, how does the DXGI_SWAP_CHAIN_DESC look like? is it in current mode or in first enumed mode (smallest dimension), i am just interested..? Nah, thats what I thought your were suggesting. As I figured that creating a swapchain requires you to set the format. I was looking for way to get the format of the users current display mode. Thx for trying anyway. any other suggestions? Or do i have to use brute force and guess work?? Raj
  4. Get the current display mode using D3D10

    I didn't think that would work as you need to fill in DXGI_SWAP_CHAIN_DESC to create the swap chain in the first place. But if i create a swapchain without filling DXGI_SWAP_CHAIN_DESC or its BufferDesc, D3D10CreateDeviceAndSwapChain will use the current display format?? Is that right? Raj
  5. Get the current display mode using D3D10

    That gives me list of possible display modes for a particular adaptor. But doesn't give me the current display mode. I could search the list for modes that match the details I get from EnumDisplaySettings, however that seems kinda roundabout way of doing things. Also to get the format you would have to store all potential formats for each bit depth and search through them. Any alternatives?? Thx Raj
  6. Hi I'm moving to D3D10 from D3D9. I know in d3d9 you can use GetAdaptorDisplayMode. But can't figure out how to do it in D3D10, specifically how to get the display format. I know that I could use EnumDisplaySettings to get the current resolution and bits per pixel, but there is no way to safely convert this to a DXGI Format. So is there a proper function to do this? Or do I need to use some sort of hack? Thx Eternal Newbie
  7. I'm trying to render a animated mesh. but I have a bug somewhere. When I try to render using the retail version of D3D9 I dont see the model. Using Pix in the render tab there is nothing, but in the mesh tab you can see the geometry rendered in the viewport. I also tried using the debug version of D3D9, with the appropriate flags set. So that I could debug the program. However under the mesh tab under PostVS and Viewport it simply says "Failed to process vertices". So I can't debug this problem. Looking at the debug output it says "Unsupported renderstate D3DRS_INDEXEDVERTEXBLENDENABLE.". Note this only occurs when using the debug version D3D9. I have tried this using my own shader and the fixed function pipeline. I have tried this with a different model in the same format. (I'm using my own max exporter for both models). This model is rendered. Obviously there are numerous things that could cause invisible geometry. But can anyone suggest why I might be getting the "Failed to process vertices" message and how I could get rid of it. So that I can debug the code? I've just installed NVidia latest perfSDK and drivers and I'm using a Gf8800GTS. if that makes any difference. Tia Eternal Newbie
  8. blooming effect without shader

    Try looking at Frame Buffer Postprocessing Effects in DOUBLE-S.T.E.A.L for an explanation on hhow to apply bloom. The way you perform a blur is by applying a convolution. This involves for each pixel\texel apply a kernel which sums neighbouring pixel with a weight. The simplest convolution is a box blur which takes all neighbouring pixel and averages them. Here the kernel might be a 3x3 where each weight 1/9th. As the kernel uniform(?) and applying a 2d kernel more expensive than applying two 1d kernel, we can optimse by applying two 1d kernels. First we apply a 3x1 kernel horizontally then we apply 1x3 kernel vertically. float hKernel[3] = {0.33f, 0.33f,0.33f}; float vKernel[3] = {0.33f, 0.33f,0.33f}; //Horizontal Blur for(int y = 0; y < height; y++){ for(int x = 1; x < width - 1); x++){ tmp = hKernel[0] * Buf[(y * width) + x - 1]; tmp += hKernel[1] * Buf[(y * width) + x]; tmp += hKernel[2] * Buf[(y * width) + x + 1]; tmpBuf[(y * width) + x) = tmp; } } //Vertical Blur for(int y = 1; y < height - 1; y++){ for(int x = 0; x < width); x++){ tmp = vKernel[0] * tmpBuf[((y - 1)* width) + x]; tmp += vKernel[1] * tmpBuf[(y * width) + x]; tmp += vKernel[2] * tmpBuf[((y - 1) * width) + x]; tmpBuf[(y * width) + x) = tmp; } } Thats off the top of my head but is roughly right. Actually the above wont work perfectly as there are details missing (like handling edge condition). But with with a little research you should be able to fill in the blanks. Also the above show a simple box blur, but are other blur techniques which may be better. (Edit) Fixed your link. <3 mittens [Edited by - mittens on July 12, 2007 8:30:27 AM]
  9. Silhouette Rendering...

    If I understand the process correctly the you can perform the operation in any space as long all the vectors are in the same space. so you Check if dot((Vertex - dir), n1) * dot((Vertex - dir), n2) < 0 Where dir will be (0,0,0) if you are in eyespace or lightspace. But its more efficient if you do the calculation in the same space. So to create a LighView matrix you would use the same techniques as for computing a View matrix. This is easiest for a spotlight as you already have a pos and lookat vector. For a directional light calcalting the vector isn't too difficult you just need to make sure it 'behind' you scene. Point light are difficult, generally the method iv seen is similar to cube mapping. Alternativly use the method i suggested above.
  10. Silhouette Rendering...

    Ok I havent tried using shadow volumes, but from what i understand you need to know if a face is front facing or back facing. For this you can use the same techniques as you would for backface culling. Which can be done by taking the sign of the dot product of the normal and the direction vector (which for culling is the view vector and for shadow volumes is the light direction). So if n1 and n2 when dotted with the dir vector have the same sign then they're not silhouette edges. This method can be used in any space.
  11. Per pixel lighting

    how long is a piece fo string? The best space todo lighting calcs depends on exactly what your are doing? which reflection model are you using? Are you doing bump mapping? etc. Also there no best solution there always a trade off between quality and performance so it depend son your specs for both, probably the main factor will be fps you want/require for a particular spec.
  12. Interpolating between normals?

    The accurate way of calculating the normal is being SLERP'ing (using Spherical interpolation) But the calculation is expensive which is why you'll almost always see it done as a linear interpolation, which is then renormalised. The formula for SLERP is:- N(t) = N0(sin(1-t)θ/sinθ) + N1(Sin(tθ)/sinθ) You can look it up on wikipedia or google if you need more detail.
  13. You can perform lighting in almost any 3d space. So the choice depends on which would be the most efficent. As your lights are already in world space this is your first option. but if your using a reflection model which uses the view vector (e.g phong) then its probably more efficient to do the calculations in eye space. Not used the fragment shader with opengl, but if you glFragcoord uses proj space then you'd have to transform your light and view vectors to proj space.
  14. Normal mapping. TBN transform

    It would be more efficient to do the transformation in the vertex shader, then the attributes would be interpolated and you would renormalise the vectors in the pixelshader. But you would lose a little accuracy. Also if you transformed the L vector to tangent space you would use a TBN matrix calculated per vertex, but as you interpolate across a polygon the interpolated tangent and normal vectors would also be different from those used in the TBN matrix so that would increase your innaccuracies. This suggests transforming to tangent space is more efficient if your using gouraud shading (not interpolating the normal) and using world space would be more efficient\accurate if you use full phong shading(interpolating the tangent and normal). Is that that right?
  15. Normal mapping. TBN transform

    For a single light using the lambert reflection model, this would mean either transformin a single normal or L vector to world or tangent space respectivly. So the only additional computation would be transposing the TBN matrix (per fragment). But if you were using another reflection model this would mean transforming one more additional vectors(e.g V, R, H, etc), and if your using multiple lights then this means (numLights * numVectorForReflModel) transformations so wouldn't it generally be more efficient to transform the normal? or am i missing something?