Extended Data Table 2 Comparison between the state-of-the-art scene flow learning methods and GotFlow3D on different flow cases in the synthetic FluidFlow3D-norm dataset

From: Recurrent graph optimal transport for learning 3D flow motion in particle tracking

  1. The best results are marked in bold. We present the performance of these methods with metrics of EPE, ACC Strict, ACC Relax, Outliers and NEPE. GotFlow3D achieves the best performance on most of the flow cases except the Uniform flow, where PV-RAFT is slightly superior to GotFlow3D on ACC Strict (Relax) as well as Outliers. In some of the complex flows (that is, forced isotropic turbulence, forced magneto-hydrodynamic turbulence and Beltrami flow) which contain small-scale flow structures, the existing flow estimators including PV-RAFT provide worse results, whereas GotFlow3D consistently performs very well.
Back to article page