Project Proposal - Flaming Cafe

Octave Arevian & Max Fruehauf

Introduction

Motivation

Image References: [1], [2] [3] [4]

The theme of this year is "Out of place": we wish to to grab the attention of the spectator and disturb them by presenting objects or phenomenons in an environment where they should not be.
We decided that we do not want the spectator to catch the oddness of the scene immediately but rather to feel more and more concerned/amused as he observes the picture thouroughly and discovers unexpected things in what seemed at first to be a very normal environment.

To achieve this goal, we first thought about finding a basic setup for our scene that would be flexible enough to add those odd elements in the scene and would still let us show off the abilities of our render. We agreed on the facade of a café seen from the road. This lets us chose a variety of atmospheres (day/night, cloudy/foggy/rainy/sunny) and gives a lot of opportunity to add odd details around and/or inside the café.

For now, even though the details are not fixed, the main source of disturbance was chosen to be a bucket of fire that would be unleashed over the café by an angry neighbour living above (The reason for this angriness is still to be determined).

Division of work

We decided on the following division of work between the two of us:

Feature Points Category Owner
Many light sampling (light acceleration data structure, volume support not necessary) 30 Advanced Max
Advanced Camera Model (e.g. depth of field, lens distortion, non-spherical aperture, chromatic aberrations; 3 effects required) 15 Medium Max
Environment Map Emitter 15 Medium Max
Heterogeneous Volumetric Participating Media (with integrator of choice, e.g. Photon Mapping or Path Tracing; includes 15 points from homogeneous) 30 Advanced Octave
Disney BSDF (5 out of 10 parameters required) 15 Medium Octave
Emissive Participating Media 10 Medium Octave
Bump Mapping or Normal Mapping 5 Basic Octave

We are set on implementing some of these features such as Heterogeneous volumes and many light sampling but are flexible on others. For instance, we are not yet decided on going for the DisneyBSDF vs. LayeredBSDF vs. MeasuredBSDF and would be happy to receive some input there on which would be best for our chosen scene. Could you also provide some advantages and disadvantages of each?

Based on the feature list above we assumed that we could need some additional features to best render our final image. We list these below with any other additional points we are planning to get during the project. We are also aware that these will not count towards the final grade but still assume them to be important enough anyway to spend time on them 😉.

Feature Points Category Owner
Modeling Meshes (counts only once) 5 Other Octave
Intel's Open Image Denoise integration 5 Other Max
Rendering using the Euler Cluster 5 Other Octave
Participating in a Group 5 Bonus Max
Participating in a Group 5 Bonus Octave
Potentially presenting at the Rendering Competition 5 Bonus Octave / Max

Feature Validation

We plan to validate most of the features listed above directly to Mitsuba 3 as suggested in the project description. For the following features we plan to setup identical scenes in nori and Mitsuba and compare their respective renders. For the others there is no direct equivalent available in Mitsuba and we propose to validate them as follows:
  • Many Light Sampling: Compare the original rendering with our Many Light Sampling and a converged reference image. As here
  • Advanced Camera Model: The following two effects are implemented in Mitsuba and we plan to compare to these directly: Depth of Field and Camera Distortion. For the non-spherical aperture we will validate it compared to a reference image with the default camera model.
  • Emissive Participating Media: Compare different sampling strategies for emissive volumes with a converged version since we did not find a suitable reference implementation online.