how to get the optical flow result

Thanks for the great work!

I have two questions about how to use it in practice:

Optical flow output
Is there a recommended way to use CoWTracker to produce optical flow-like results (like CoWTracker paper in **Optical Flow**)?
For example, can the predicted point tracks be converted into dense or semi-dense optical flow, or is CoWTracker mainly intended for sparse point tracking rather than optical flow estimation?

Online / streaming mode
Can CoWTracker be used in an online setting, where frames arrive one by one and point tracks are updated incrementally?
In other words, is it possible to run CoWTracker frame-by-frame for real-time or streaming applications, instead of processing the full video clip offline?

Any suggestions or recommended usage patterns would be very helpful. Thanks!

The issue asks about CoWTracker's optical-flow output and online frame-by-frame use; start by locating the repository's public inference entry point and any usage documentation or paper guidance. Done means documenting whether point tracks support dense or semi-dense flow and whether incremental streaming is supported, with clear usage patterns.

Written by the indexing model from the issue text.