ICMLCN2025 - Towards Generative Ray Path Sampling for Faster Point-to-Point Ray Tracing
Posters and 3-minute thesis presented at ICMLCN 2025!
At ICMLCN 2025, I presented our paper Towards Generative Ray Path Sampling for Faster Point-to-Point Ray Tracing (missing reference). This work is the result of a collaboration with Pr. Degli-Esposti’s laboratory from University of Bologna, where I worked for 4 months in late 2024. In this work, we introduce a novel Machine Learning approach to Ray Tracing: sampling path candidates using a generative model. The core motivation is to avoid the exponential complexity of generating all path candidates by using a surrogate model that learns how to only suggest the most promising candidate rays, i.e., the candidates are likely to generate a physically valid ray path.
My participation to ICMLCN was also a great opportunity to demonstrate DiffeRT, the open-source Ray Tracing tool I develop for my research, so I prepared a small poster for that too. I also participated in the on-site 3-minute thesis contest.
Finally, an interactive tutorial is available to guide the readers through the implementation and training of our presented model.
Media
Below, you can find the different media I used to present my research at ICMLCN.
Paper poster
Demo poster
I also made a small 1-minute video showcasing DiffeRT’s main features.
3MT video
Unfortunately, the on-site contest was not recorded, but you can find the video I recorded when submitting my participation to the contest.