The intention of this project was to create a convincing interior visualization scene of a high-end, modern hotel suite. This scene makes use of plants made during my Project: Plants work.
This scene also consists of very difficult to resolve light paths, with all light coming through the open doors, and makes use of Intel's Open Image Denoise library which is included with LuxCore, and which will be discussed further.
LuxCoreRender 2.5 & Blender 2.92
One of the most significant challenges with creating an entire room (that is to say, a space with all four walls present)is to maintain a sense of scale while modelling each item. To do this, an accurately sized "mannequin" was placed around the scene, and beside each object during its creation, in order to ensure the room conformed to normal, human proportions.
As mentioned previously, all light in the scene is provided by a Sun & Sky Lamp and enters through the open glass doors at the far side of the room. Despite not using image based lighting, like HDRI, the Global Illumination caused by glossy surfaces gives the light a rich, warm look.
Each frame presented here was rendered at an extremely high 4704x3136 pixels; higher than 4K resolution. Despite this, the average render time was only about 15 minutes on a single-processor system. This is because LuxCoreRender includes, for optional use, Intel's Open Image Denoise. This library makes use of the scene information to "intelligently" smooth out and reduce noise. While renders appear extremely - in many cases unnaturally - clean, the trade-off is that a substantial amount of detail is lost.
Below I have compiled a comparison of the scene with OIDN enabled, with it disabled, and when rendered with LuxRender 1.6.
The comparison highlights some key differences in the behavior of each batch of settings. It is clear from the "denoising-disabled" set of images that LuxRender 1.6 gains the upper hand with difficult light paths.
Difficult light paths, in this case, include the transmission of light through volumes (like glass) and the parts of the scene obscured in shadow (which do not recieve direct light, rather have light bounced onto them by directly illuminated surfaces). Metropolis Light. Transport (MLT) is used with the Bidirectional integrators, which "guides" computation towards certain high-contribution light paths in order to better sample relevant areas of the iamge. These paths are then mutated slightly, resulting in faster overall convergence due to the focus around high-contribution areas. LuxRender 1.6 also has the benefit of enabled pixel filtering, which blurs the image slightly in order to transfer energy from converged areas across the image plane. This leads to lower percieved "noise" as dark spots are smoothed over.
In terms of overall cleanliness, the OIDN images are the clear choice, however some amount of sharpness and detail is lost in these images.
At right are some images of the scene in the 3D viewport, prior to rendering. Each piece of gray geometry shown here has certain surface properties applied which are computed when the final image is rendered, resulting in the wide array of materials seen in the above images.