@janusch_patas
Gaussians on Fire: High-Frequency Reconstruction of Flames Contributions: 1. We introduce Gaussians on Fire, a complete pipeline for reconstructing spatio-temporal flame dynamics from only three synchronized cameras. 2. We propose a flow-initialized dynamic Gaussian representation, where dense optical flow fields are fused across views to form an initial 3D flow volume that drives Gaussian motion initialization. 3. We construct and evaluate a real-world dataset of fire sequences captured with a hardware-synchronized three-camera rig and present quantitative and qualitative experiments demonstrating that our approach produces temporally coherent, high-frequency reconstructions of flame dynamics under sparse capture conditions. 4. We define a capturing setup that only requires three cameras and a hardware synchronization pattern that allows us to utilize rolling shutter artifacts for microsecond-level synchronization of the cameras.
