sfpick performs automatic picking from semblance-like panels.
The underlying algorithm is described in Appendix B of the paper Velocity analysis using AB semblance and is inspired by the method of virtual endoscopy in medical imaging. The following example from tccs/avo/avo shows an automatic pick through AB semblance.
If smooth=n, the program picks an optimal path through the panel but does not try to smooth it. If smooth=y (the default), the picked path is additionally smoothed using shaping regularization. The smoothing radius across different dimensions is controled by rect1=, rect2=, etc. parameters. The number of shaping iterations is specified by niter=. The following example from sep/vc2/beivc shows two different picks for two different values of the rect1= parameter
The vel0= refers to the value the pick initially takes at the surface (the origin). This value is not necessarily preserved, because it can get modified by smoothing. In the following example from tccs/timelapse/timelapse, vel0=1.
The an= parameter controls the anisotropy between the two axes of the semblance panel. Varying this parameter, you can achieve picks that are more rigid or more flexible. The gate= parameter controls how far (in velocity samples) the pick can swing between two neighboring time samples. Changing its value also affects the flexibility of the picked trend.
For picking from 3-D (rather than 2-D) panels, try sfpick3.