Neural UpFlow

We present a novel up-resing technique for generating high-resolution liquids based on scene flow estimation using deep neural networks. Our approach infers and synthesizes small- large-scale details solely from low-resolution particle-based liquid simulation. The proposed network leverages neighborhood contributions to encode inherent properties throughout convolutions. also propose interpolate between generated varying simulation discretizations state-of-the-art bidirectional optical solver method fluids in addition key-event topological alignment constraint. In conjunction with the contributions, our loss formulation allows inference model epochs reward important differences regard significant gaps discretizations. Even when applied an untested setup, is able generate plausible details. Using this interpolation predicted displacements, combines input motion infer semi-Lagrangian advection. furthermore showcase how can facilitate large datasets subset of initial condition parameters.