Archived

This topic is now archived and is closed to further replies.

An Anisotropic Phong BRDF Model

This topic is 5143 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

I'm writing a raytracer and are trying to replace the simple Phong model with the paper "An Anisotropic Phong BRDF Model" by Ashikhmin and Shirley. Maybe someone could help me with a few questions. 1. I don't need the model to be anisotropic. How should I set the u and v vectors? 2. When setting nu and nv to the samples in the paper 10-10000 I get HUGE values in the exponation. So I set them between 0-1 and it looks a bit better. What is correct here? 3. How should I set the Rs and Rd colors? Can I set them totally independent? 4. The n,h dot product sometimes gives me a negative value and I then get a domain error for the pow() function. How should I handle this? 5. Another model I'm thinking of is the Strauss model from "Principles of digital image synthesis" by Glassner. Has anyone implemented this one? Cheers [edited by - Patrik Svensson on November 12, 2003 5:36:41 AM]

Share this post


Link to post
Share on other sites
If U and V are the same then it is no longer anisotropic.

The nice thing about the Ashikmin Shirley model is that U and V are very intuitive - the U represents the specular power across the U axis, and the V represents the specular power across the V access (defined by your 2 tangent vectors).

If you are trying to represnet Metals, then you should throw away the Schlicks aproximation of Fresnel Term for the specular highlight and basically get rid of the Diffuse term.

Othwerise, if U and V are set to the same thing you end up with a Phong model which more correctly models dialetric models by compensating for the varying thickness of the semitranslucent layer at grazing angles to the light and view.

Rs and Rd are a little counter Intutive. If I remember correctly, Rd = amount of diffuse that wasn''t absorbed by the Specular dialetric layer. So, if Rs = 1.0 nothing would even get to the diffuse layer. Basically, the perceived diffuse = (1-Rs)*Rd.



Share this post


Link to post
Share on other sites
>I suggest the Torrance-Sparrow model; it is very simple
>to implement and quite powerful, without some of
>Phong''s physical inaccuracies.

Don''t be ridiculous. If a graphics-god like Peter Shirley co-derived the BRDF in the year 2000, how can it even cross your mind that the model would be physically inaccurate?

Lewis derived the physically accurate Phong BRDF back in 1994, btw.

- Mikko Kauppila

Share this post


Link to post
Share on other sites
Take a large planar surface, place the camera slightly above it, and put a light source at a large distance away at similar height to the camera. Compare Phong to Torrance-Sparrow; Phong does not correctly depict the specular highlight''s long, narrow shape. Torrance-Sparrow is at least based on empirical lighting data, as opposed to Phong, which (in every single place I''ve seen it mentioned with regard to other models) is entirely arbitrary and designed for good appearance, not physical accuracy.


By the way, Lewis'' revision of Phong is not Phong, it is a separate, revised model. I''ve never heard anyone refer to his model as pure Phong, always qualified as his particular modification. I see no indication in this thread that we''re talking about modified Phong.

Share this post


Link to post
Share on other sites