Verity and Performance

Feb 23, 2011 at 7:55 AM

Depending on applications, there are different requirements to choose between speed and quality.

1. For egomotion and traffic following, the speed is crucial but the missing small objects after the edge detection is not a problem.

2. For most other purposes, the goal is to make the edge detection just so good as the human can do.  Every visible object, big and small, should be outlined and there should be no false or duplicate edges (verity). The speed is also an important factor but of the second priority.  The best way of increasing speed without the quality sacrifice, is ubiquitous caching of calculations. 

3. The scientific and forensic application may require the utmost quality, better than human can do.  This task is deferred to distant future, there should be developed the criteria how to measure that high quality.

The Outliner project is focused on the type (2) applications. The calculation of the curvature will improve verity at the expense of performance - I think this is an acceptabe trade-off.