SurfPhase: 3D Interfacial Dynamics in Two-Phase Flows from Sparse Videos

TL;DR

This work tackles the challenge of reconstructing 3D interfacial dynamics in two-phase flows from sparse-view videos. It introduces SurfPhase, a two-stage pipeline that represents the liquid-vapor interface with dynamic Gaussian surfels augmented by a signed distance function for geometric consistency and uses diffusion-based priors to refine novel-view renderings from limited camera coverage. Bubble-guided velocity estimation ties surfels to individual bubbles to enable metric velocity recovery, improving temporal coherence and physical plausibility. The authors collect a pool-boiling dataset with monocular videos for diffusion training and synchronized dual-view data with metric calibration, plus synthetic scenes for ground-truth evaluation, and demonstrate superior novel-view synthesis, geometry reconstruction, and velocity estimation compared with baselines.

Abstract

Interfacial dynamics in two-phase flows govern momentum, heat, and mass transfer, yet remain difficult to measure experimentally. Classical techniques face intrinsic limitations near moving interfaces, while existing neural rendering methods target single-phase flows with diffuse boundaries and cannot handle sharp, deformable liquid-vapor interfaces. We propose SurfPhase, a novel model for reconstructing 3D interfacial dynamics from sparse camera views. Our approach integrates dynamic Gaussian surfels with a signed distance function formulation for geometric consistency, and leverages a video diffusion model to synthesize novel-view videos to refine reconstruction from sparse observations. We evaluate on a new dataset of high-speed pool boiling videos, demonstrating high-quality view synthesis and velocity estimation from only two camera views. Project website: https://yuegao.me/SurfPhase.

Figures (12)

• Figure 1: We introduce a new task: Reconstructing 3D interfacial dynamics from sparse-view videos.
• Figure 2: Overview. Given input videos of two-phase flows, SurfPhase reconstructs the 3D appearance, geometry, and velocity. During the refined reconstruction stage, we use same geometric surface constraint and bubble guidance as in initial reconstruction. We summarize the pipeline in Alg. \ref{['algo:dyn_bubble_velocity']} in the appendix.
• Figure 3: Our real two-phase flow pool boiling setup (left) and samples of our calibration image pairs (right).
• Figure 4: Novel view video synthesis on captured real data.
• Figure 5: Novel view video synthesis on synthetic data.