Convection Over Coffee: Revisiting a Steamy Background-Oriented Schlieren Dataset

“Flow over an espresso cup” is a modern classic in the background-oriented schlieren (BOS) literature, demonstrating that three-dimensional, time-resolved velocity, pressure, and temperature fields can be reconstructed from multi-camera BOS measurements. In that study, a buoyancy-driven flow above a heated cup was measured using BOS tomography, and density reconstructions were subsequently post-processed with a physics-informed neural network based on the Boussinesq approximation of the Navier–Stokes equations. In the present work, we revisit this canonical dataset using the neural-implicit reconstruction technique in both standard and physics-informed configurations, enabling a direct assessment of the benefits and limitations of physical constraints in the reconstruction workflow. Comparisons are made between historical results and modern best practices. The influence of temporal resolution on the inferred flow fields is systematically evaluated.