Hi,

I'm working on an inverse rendering project and we're using Radiance to
generate synthetic data for validation. Given a scene with diffuse
surfaces, we're using the radiosity formulation as the core of our inverse
process, namely that at any given scene point:

radiosity at point = (reflectance at point) x
sum_over_scene_points(radiosity at scene point x form factor between points)

The sum is calculated using a hemicube method (with suitably high
resolution) and the radiosities are obtained from reprojecting the .hdr
outputs of Radiance onto the scene geometry.

My problem is that even in simple box scenes this equation doesn't hold.
Most of the time, the reflectance estimated from this equation is too low.
I've used enough ambient bounces that the Radiance output has converged,
but I'm wondering if there are other Radiance rendering options that might
be causing these problems.

Edward

Edward,

When I create ground truth renderings I pay close attention to the
following parameters:
-aa 0
-ad very high (like 10,000 - 50,000)

Also, if you're using large area illumination sources, use the glow
material instead of light material. If you're use light material for small
area sources pay attention to the direct sampling parameters (-d?).

Andy

···

On Wed, Sep 24, 2014 at 3:04 PM, Edward Zhang <[email protected]> wrote:

Hi,

I'm working on an inverse rendering project and we're using Radiance to
generate synthetic data for validation. Given a scene with diffuse
surfaces, we're using the radiosity formulation as the core of our inverse
process, namely that at any given scene point:

radiosity at point = (reflectance at point) x
sum_over_scene_points(radiosity at scene point x form factor between points)

The sum is calculated using a hemicube method (with suitably high
resolution) and the radiosities are obtained from reprojecting the .hdr
outputs of Radiance onto the scene geometry.

My problem is that even in simple box scenes this equation doesn't hold.
Most of the time, the reflectance estimated from this equation is too low.
I've used enough ambient bounces that the Radiance output has converged,
but I'm wondering if there are other Radiance rendering options that might
be causing these problems.

Edward

_______________________________________________
[email protected]

Suggest using -dt 0 if you are computing many ambient bounces.

-Greg

···

From: Andrew McNeil <[email protected]>
Date: September 25, 2014 10:25:21 AM PDT

Edward,

When I create ground truth renderings I pay close attention to the following parameters:
-aa 0
-ad very high (like 10,000 - 50,000)

Also, if you're using large area illumination sources, use the glow material instead of light material. If you're use light material for small area sources pay attention to the direct sampling parameters (-d?).

Andy

On Wed, Sep 24, 2014 at 3:04 PM, Edward Zhang <[email protected]> wrote:
Hi,

I'm working on an inverse rendering project and we're using Radiance to generate synthetic data for validation. Given a scene with diffuse surfaces, we're using the radiosity formulation as the core of our inverse process, namely that at any given scene point:

radiosity at point = (reflectance at point) x sum_over_scene_points(radiosity at scene point x form factor between points)

The sum is calculated using a hemicube method (with suitably high resolution) and the radiosities are obtained from reprojecting the .hdr outputs of Radiance onto the scene geometry.

My problem is that even in simple box scenes this equation doesn't hold. Most of the time, the reflectance estimated from this equation is too low. I've used enough ambient bounces that the Radiance output has converged, but I'm wondering if there are other Radiance rendering options that might be causing these problems.