an0malaus opened this issue on Jan 29, 2020 ยท 3 posts
an0malaus posted Wed, 29 January 2020 at 2:16 PM
If you please?
As I render out a series of still frames, to the limit of both my abilities and patience (due to avoidance of inordinate render times), I seem to achieve a blend of both success and failure in equal parts. I can set a scene, morph props and figures to populate it, craft new materials aided by the secrets gleaned from decades of browsing the forums where the masters deign to cast their pearls of wisdom, and yet, I keep encountering new (to me) problems that I do not yet have a solution for.
I'm choosing to render exclusively with SuperFly in Poser, due to my hardware's age and incompatibility with other, external, GPU capable render engines, so any suggestion to render elsewhere will be completely off-topic. This is about Poser's capabilities.
This scene starts with an environment sphere and HDR background image. All such scenes have inherent restrictions on where cameras within the sphere can be placed and match their optical parameters to blend figures and props believably with the background. I have a pool floor and matching water prop in place of a ground plane, and get reasonable success with volumetric absorption and scatter. I can see that some texture matching will blend the series of tree props better with the background, though there's no real help for the environment image including essentially foreground objects like the ground cover plants, except hiding it with more props. So far, all relatively fixable, with wisely chosen camera framing.
The next render in the sequence does most of what I'm looking for, with a wave deformer adding some realism to the reflections and refraction. More problems at the shore-line are purely due to an inability to warp the environment sphere image to match a viewpoint that doesn't exactly correspond to where the image was acquired. Perhaps playing with the tree props will hide the fact that they haven't appeared to move in tandem with the foreground components of the environment image. The tree, however, are in the correct place relative to the camera, so hiding more of the background is the only solution (that I'm aware of) there.
Moving on:
This render looks like a total disaster, but it really only has a few glaring problems that need fixing. The figure is in the same pose and position as the previous render, with the camera submerged within the enclosed volume of the water prop, which intersects the pool bottom prop so there are no surface reflections from the bottom surface of the water. The fireflies will go away if I'm prepared to wait till the heat death of the universe for renders to complete, so I ignore them for the moment.
The rippled water surface plays merry hell with any expectations of clearly visible reflections from below and refraction from above, but it's all there, just messed about by the angles of incidence. All good and realistic, in the main. Apart from a dark & murky appearance due to excessive settings of volumetric absorption and scatter densities, which I've subsequently adjusted in more recent renders, the glaring problems are the dark artifacts, which don't obviously align with or correspond to visible items in the scene.
The figures involved are V4, V4Leotard and all the corresponding anatomical figures designed for V4: Skeleton, Breasts, Gut & Thorax. The Skeleton has opaque materials, but the organs are all using essentially the same volumetric material nodes as the pool water, including the same IOR, so in that sense, I wouldn't expect a lot of artifacts. However, I've seen circumstances where volumetrics don't always play nicely if objects are not manifold (entirely enclosed volumes with no edges belonging to only one facet). Essentially, the depth of the ray going to infinity, due to not finding a back face of the volume to reflect from means the absorption goes to its maximum and you get black.
Most unfortunately, that infinite absorption happens to the ray directly from the camera to the surface of the open volume, so it doesn't get refracted any more, meaning the artifacts are in direct line of sight of the original object/ray intersection, and not the refracted position implied by passing through the water's surface. I'll call them Neutrino Artifacts, since they don't seem to interact with anything, anymore.
So, if open volumes are causing these artifacts, why are they open? The dark "moustache" artifact at the bottom, happens to correspond with the weld between the hip and abdomen actors of the Gut figure. Hiding the gut figure makes most of the artifacts go away, but the V4 figure's skin has transparency and volumetrics applied to it too, yet it's not generating artifacts. What's different?
It turns out V4 has Unimesh skinning and subdivision. Changing the organ figures all to unimesh skinning made almost all of the artifacts go away, but frustratingly, not all. Why? Hiding individual actors and doing area renders of the remaining artifacts revealed the worst culprit: the neck. What's up with that? Well, it turns out to have an open hole, which is the top of the oesophagus, so of course, any rays passing in through one side and out the open top will proceed to infinity without meeting another face which closes that volume, so black!. If I hide the neck actor, that just moves the problem further down the oesophagus to the top of the chest actor, which then has an open hole generating a new artifact.
I may just have to bite the bullet and model a surface to close the top of the oesophagus, make it a neck2 actor and weld it to the neck.
Before resolving (partly) those problems, and then having a render rudely interrupted by a power outage while I was sleeping, I managed a raytrace preview of the next frame in the sequence, which, being fully submerged, avoids all the distortions of the rippled surface, and having adjusted the absorption and scatter densities to enhance the visibility, gave me this:
Still all the original artifacts present, but much clearer view, so, artifacts mostly sorted, here's progress on the current render of the partially submerged scene:
And finally, my question for the node masters becomes: Is there some shenanigans with ray depth parity that could be performed to cull the infinite absorption depth rays?
P.S. here are my render settings. When I bump the samples from 5 (25) to 7 (49) it takes twice as long (expectedly) and removes some of the fireflies.
Verbosity: Profusely promulgating Graham's number epics of complete and utter verbiage by the metric monkey barrel.
an0malaus posted Wed, 29 January 2020 at 5:12 PM
Re-render completed. Most artifacts eliminated.
Volumetric absorption and scatter densities set to 25% of original value.
Verbosity: Profusely promulgating Graham's number epics of complete and utter verbiage by the metric monkey barrel.
an0malaus posted Thu, 30 January 2020 at 2:44 PM
And a full render with most of the previous artifacts suppressed:
Verbosity: Profusely promulgating Graham's number epics of complete and utter verbiage by the metric monkey barrel.