Panoramic Camera Awesome

[treddie]

After doing some tests and looking at the geometry, there is no perfect solution when setting the PanCamera's origin.

In a scene with "real" emitters (mesh emitters), the subject is located with respect to those emitters. So, imagine standing inside the subject and looking outward, with your eyes at the centroid of the subject . Out there in the Octane world are three mesh emitters. As you turn around on your feet by some angle, (ALPHA) you find that you are facing one of those emitters. Not only is the emitter at angle ALPHA relative to the direction you were originally pointing, but it also has a given elevation angle (BETA), above or below your eye level. As you continue to turn, you will find the other two emitters at their own ALPHA and BETA angles. Technically, your eye position, the centroid of the subject, should be the PanCamera's position. And in order for the emitters to be baked into their proper position in the HDRI, you also need to have the PanCamera's target set parallel to Octane's world, -Z axis. For instance, if the PanCam's origin is at (1, .5, -3), it's target needs to be at (1, .5, -4). Note that the target (z) value is further down the -Z direction than the origin's (z) value. This insures that the projection of those emitters out onto the HDRI sphere (or cylinder) will be accurately placed relative to the subject. Now, remove everything from the scene except the emitters and render the scene. Save out the render as an Untonemapped .exr image. This is your HDRI lighting environment image.

In an ideal world, when this HDRI is used as the replacement for the mesh emitters in the original scene, all cast shadows and reflections SHOULD be accurate. Unfortunately, this is not entirely true. The lighting on all diffuse surfaces, as well as cast shadows will be accurate, PROVIDED THE SUBJECT IS COMPARABLY SMALL RELATIVE TO ITS DISTANCE FROM THE EMITTERS. However, any REFLECTIONS in glossy or specular surfaces will NOT be so accurate. The reason is due to parallax shift...The scene camera is NOT at the center of the subject, therefore does not see the HDRI emitters in the exact same positions relative to their mesh emitter counterparts. Remember, the mesh emitters were physical objects somewhere inside the 3D scene's space. When those emitters are baked into the HDRI, they now are located OUTSIDE the scene, out on the surface of the HDRI sphere (or cylinder). Therefore, the reflections of those emitters will not appear in the same positions compared to the scene being lit by real mesh emitters. The only case where parallax shift would not occur would be if the scene camera's origin was ALSO at the subject's centroid, which means you wouldn't see the subject anymore! The only scenario where these parallax errors can be minimized is if your original mesh emitters are so far away, that the camera sees them as being in about the same position as they appear in the equivalent HDRI. Sun light is a perfect example in this respect. The Sun is, for all intents and purposes, infinitely far away. It will also be, for all intents and purposes, when burned into an HDRI emitter.

Keep in mind that this parallax error can show up quite dramatically in textures such as brushed metal. Here, anisotropy in the textures, is really acting like zillions of individual reflective surfaces, each reflecting back their own reflection of the light sources. So a slight parallax error that might show up as a slight shift of a single surface's reflection, will be multiplied for every anisotropy reflection in the brushed metal texture.

I suppose this is where the power of HDRLightStudio comes in (Thanks to, dionysiusmarquis in my other thread, "Here's a Thought", for providing the link to this program). With that program, you can see in real time exactly what the HDRI emitters will do as you move them into position. This capability would be a fine addition to Octane as well...Unless we replace this entire scheme with procedural emitters with IES, that can be placed anywhere we want them, nixing the idea of HDR altogether.

However, if you can live with the somewhat different state of the reflections when replacing mesh emitters with their baked HDRI proxy, the savings in render time can be massive, and just as spectacular looking.

Of course, if you are only working with diffuse surfaces, there should be no differences at all, if the subject is comparably small relative to its distance from the emitters.