Supervised Learning
The supervised learning setting has been intensively studied. It takes pairs of velocity maps and seismic data as input and output during training. The seismic data is usually generated by forward modelling from the velocity maps. OpenFWI provides 6 datasets for this setting: FlatVEL, CurvedVel, Kimberlina CO2 Leakage, Style-transfer and Kimberlina 3D.