🔒 Locked

Image Space Motion Blur Algorithm

Started by luridcortex Apr 16, 2006 at 5:54 PM 3 replies 4.9k views

Original Post

Apr 16, 2006 at 5:54 PM

Im trying to implement the motion blur algorithm described in the paper "Stupid OpenGL Shader Tricks" -> http://developer.nvidia.com/docs/IO/8230/GDC2003_OpenGLShaderTricks.pdf by Simon Green. Im having a some trouble visualizing the order of operations in the algorithm. In the slides it goes as such: 1. Render current scene to texture 2. Calculate velocity at each pixel 3. Render motion blurred scene What I am doing is: 1. Rendering scene to texture 2. Render blurred scene to texture a. Calculate velocity from prev frame to cur frame in vertex shader b. Sample prev render frame and cur render frame in frag shader 3. Blend blurred scene over cur render frame Im just getting a little confused as to whether the final render should be applied as a post process effect or not. I guess you would have to split it up into two steps and have the fragment shader write the velocity to a texture?

joanusdmentia

Apr 16, 2006 at 6:05 PM

While I haven't looked at the paper, I would have thought you'd be able to perform steps 2 & 3 in one fragment shader. You're already accessing the current frame's texture in step 2 so why not perform the blend (which I'm presuming is just a normal pixel-by-pixel blend) at the same time?

luridcortex

Apr 16, 2006 at 6:21 PM

Yeah, that is how I am doing it. I had just confused myself with how the blur would be applied 'outside' of the image. But thats what the whole geometry stretching deal solves... even tho I dont see this effect in the nvidia demo.

OrangyTang

Apr 16, 2006 at 6:40 PM

Quote:
Original post by luridcortex
What I am doing is:

1. Rendering scene to texture
2. Render blurred scene to texture

a. Calculate velocity from prev frame to cur frame in vertex shader
b. Sample prev render frame and cur render frame in frag shader

3. Blend blurred scene over cur render frame

You can't just blur the scene beforehand (your step 2) as the velocity is supposed to affect the direction of the blur at that pixel.

Step 2 and 3 (calc velocity and add blur) should be done in one pass. The vertex shader extrudes the mesh and generates the velocity as the output colour. The pixel shader then uses this velocity to take multiple samples *of the original scene* to generate the directional blur for this pixel.

luridcortex

Apr 16, 2006 at 6:56 PM

I meant a/b to part of step 2, a little confusing, sorry!

Thanks for the explanation OrangyTang :)

The main problem I am having now is with calculating the previous vertex position. I was using the trick in the demo to transform the incoming vertex by the previous modelview matrix. However I am thinking this wont work in my case since I am testing with an animated model using vertex animation.

edit: Problem solved, maybe! Anyway, I just wasnt getting the correct modelview from the application - the above formula should work I think.

[Edited by - luridcortex on April 16, 2006 9:56:27 PM]

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.