OpenGL/OpenTK - Multiple Framebuffers

Question

I want to do two post processing steps for my edge-detection algorithm: 1. Find edges via Sobel-Edge detection algorithm 2. Thin the edges via morphological thinning

So what I want to achieve is to render the whole scene into a separate framebuffer, sample the created texture to find all edges and finally sample the newly generated texture again to thin the edges.

So I assumed that I need three framebuffers (one for edge detection, one for thinning and the default framebuffer) to achieve this task. The first additional framebuffer uses three textures, namely for color, normals and depth. The second additional framebuffer should use the generated images of the first framebuffer to do the thinning and output it again.

If I use two framebuffers (default fb and edge detection fb) everything works fine. But as soon as I add an additional third fb, the texture generated by this third framebuffer is not displayed. Instead a black window is displayed.

So here is the code for the initialization of the first framebuffer:

private void GenerateFramebuffer()
    {
        //Generate framebuffer
        framebuffer = GL.GenFramebuffer();
        GL.BindFramebuffer(FramebufferTarget.Framebuffer, framebuffer);

        // create a RGBA color texture
        GL.GenTextures(1, out textureColorBuffer);
        GL.BindTexture(TextureTarget.Texture2D, textureColorBuffer);
        GL.TexImage2D(TextureTarget.Texture2D, 0, PixelInternalFormat.Rgba,
                            glControl1.Width, glControl1.Height,
                            0, (PixelFormat)PixelInternalFormat.Rgba, PixelType.UnsignedByte,
                            IntPtr.Zero);

        GL.TexParameter(TextureTarget.Texture2D, TextureParameterName.TextureMinFilter, (int)TextureMinFilter.Linear);
        GL.TexParameter(TextureTarget.Texture2D, TextureParameterName.TextureMagFilter, (int)TextureMinFilter.Linear);
        GL.BindTexture(TextureTarget.Texture2D, 0);

        // create a RGBA color texture for normals
        ... generated like texture above

        // create a depth texture
        ... generated like texture above

        ////Create color attachment texture
        GL.FramebufferTexture(FramebufferTarget.Framebuffer, FramebufferAttachment.ColorAttachment0, textureColorBuffer, 0);
        GL.FramebufferTexture(FramebufferTarget.Framebuffer, FramebufferAttachment.ColorAttachment1, normalBuffer, 0);
        GL.FramebufferTexture(FramebufferTarget.Framebuffer, FramebufferAttachment.DepthAttachment, depthBuffer, 0);

        DrawBuffersEnum[] bufs = new DrawBuffersEnum[2] { (DrawBuffersEnum)FramebufferAttachment.ColorAttachment0, (DrawBuffersEnum)FramebufferAttachment.ColorAttachment1 };
        GL.DrawBuffers(bufs.Length, bufs);

        GL.BindFramebuffer(FramebufferTarget.Framebuffer, 0);
    }

Everything works well, no problems appear. The second framebuffer is created equally, except that I need only one color attachment:

  private void GenerateFramebuffer2()
    {
        //Generate framebuffer
        framebuffer2 = GL.GenFramebuffer();
        GL.BindFramebuffer(FramebufferTarget.Framebuffer, framebuffer2);

        // create a RGBA color texture
        GL.GenTextures(1, out edgeBuffer);
        GL.BindTexture(TextureTarget.Texture2D, edgeBuffer);
        GL.TexImage2D(TextureTarget.Texture2D, 0, PixelInternalFormat.Rgba,
                            glControl1.Width, glControl1.Height,
                            0, (PixelFormat)PixelInternalFormat.Rgba, PixelType.UnsignedByte,
                            IntPtr.Zero);

        GL.TexParameter(TextureTarget.Texture2D, TextureParameterName.TextureMinFilter, (int)TextureMinFilter.Linear);
        GL.TexParameter(TextureTarget.Texture2D, TextureParameterName.TextureMagFilter, (int)TextureMinFilter.Linear);
        GL.BindTexture(TextureTarget.Texture2D, 0);

        ////Create color attachment texture
        GL.FramebufferTexture(FramebufferTarget.Framebuffer, FramebufferAttachment.ColorAttachment0, edgeBuffer, 0);

        DrawBuffersEnum[] bufs = new DrawBuffersEnum[1] { (DrawBuffersEnum)FramebufferAttachment.ColorAttachment0};
        GL.DrawBuffers(bufs.Length, bufs);

        //No need for renderbuffer object since we don't need stencil buffer yet

        GL.BindFramebuffer(FramebufferTarget.Framebuffer, 0);
    }

And now comes the rendering part: First I bind the first framebuffer and use the fragment shader to output normals and color values to the two attachments:

  GL.BindFramebuffer(FramebufferTarget.Framebuffer, framebuffer);

  GL.Clear(ClearBufferMask.ColorBufferBit | ClearBufferMask.DepthBufferBit); // We're not using stencil buffer so why bother with clearing?            

  GL.Enable(EnableCap.DepthTest);
  Shader.Use();
  Model.Draw();
  GL.BindFramebuffer(FramebufferTarget.Framebuffer, 0);

Then I bind the second fb and use the textures that the previous framebuffer generated:

 GL.BindFramebuffer(FramebufferTarget.Framebuffer, framebuffer2);
 GL.Clear(ClearBufferMask.ColorBufferBit | ClearBufferMask.DepthBufferBit); // We're not using stencil buffer so why bother with clearing?  
 //Find edges and put them in separate texture (colors not needed here)
 edgeDetectionShader.Use();
 GL.ActiveTexture(TextureUnit.Texture1);
 GL.BindTexture(TextureTarget.Texture2D, normalBuffer);
 GL.Uniform1(GL.GetUniformLocation(edgeDetectionShader.program, "normalTexture"), 1);
 GL.ActiveTexture(TextureUnit.Texture2);
 GL.BindTexture(TextureTarget.Texture2D, depthBuffer);
 GL.Uniform1(GL.GetUniformLocation(edgeDetectionShader.program, "depthTexture"), 2);

And finally I bind to the default framebuffer again and draw the generated textures onto a simple quad:

 /////////////////////////////////////////////////////
 // Bind to default framebuffer again and draw the 
 // quad plane with attched screen texture.
 // //////////////////////////////////////////////////
 GL.BindFramebuffer(FramebufferTarget.Framebuffer, 0);
 // Clear all relevant buffers
 GL.ClearColor(1, 1, 1, 1); // Set clear color to white 
 GL.Clear(ClearBufferMask.ColorBufferBit);
 GL.Disable(EnableCap.DepthTest); // We don't care about depth information when rendering a single quad

 //Combine edges with color values
 finalImageProzessingShader.Use();
 GL.BindVertexArray(quadVAO);

 //Testing purpose: Send color texture to the shader
 GL.ActiveTexture(TextureUnit.Texture0);
 GL.BindTexture(TextureTarget.Texture2D, textureColorBuffer);
 GL.Uniform1(GL.GetUniformLocation(finalImageProzessingShader.program, "screenTexture"), 0);

 //Testing purpose: Send normal texture to the shader
 GL.ActiveTexture(TextureUnit.Texture1);
 GL.BindTexture(TextureTarget.Texture2D, normalBuffer);
 GL.Uniform1(GL.GetUniformLocation(finalImageProzessingShader.program, "normalTexture"), 1);

//Testing purpose: Send depth texture to the shader
GL.ActiveTexture(TextureUnit.Texture3);
GL.BindTexture(TextureTarget.Texture2D, depthBuffer);
GL.Uniform1(GL.GetUniformLocation(finalImageProzessingShader.program, "depthTexture"), 2);

//Send the texture of fb2 to shader (generates black screen)
GL.ActiveTexture(TextureUnit.Texture0);
GL.BindTexture(TextureTarget.Texture2D, edgeBuffer);
GL.Uniform1(GL.GetUniformLocation(finalImageProzessingShader.program, "depthTexture"), 3);

GL.DrawArrays(PrimitiveType.Triangles, 0, 6);
GL.BindVertexArray(0);

So outputting color texture, depth texture or normal texture generated by framebuffer one works great, but if I output the texture generated by framebuffer two, I get a black screen. I also tried it with TextureUnit.Texture4 when using edgeBuffer, but this doesn't work either.

Is this even the right approach when I want to do 2 post-processing steps?

enne87 enne87 · Accepted Answer · 2016-08-03T20:09:13

Well, the answer is to render the scene to the quad, although it is not the final image that should be output to the display.

So the steps for perform multipe post processes are:

-> Bind 1st render texture
-> Draw scene

-> Bind 2nd render texture
-> Feed 1st render texture to a shader and draw quad

-> Bind 1st render texture
-> Feed 2nd render texture to another shader and draw quad

-> Bind 2nd render texture
-> Feed 1st render texture to a third shader and draw quad

...
-> Unbind
-> Draw quad

(see Post processing and resulting texture)

So the above code is correct, only the drawing to the quad after the first framebuffer is missing (just in case anyone else needs this too).

OpenGL/OpenTK - Multiple Framebuffers

1 Answers