About Parameters

About the parameters

1. Regularization and Momentum factors

This experiments are done in parking_lot high sequence of VIODE dataset.

Using viode_stereo.launch, you can get the following result depending on the parameters (ATE in [m]):

2. Convergence rate threshold

alternating_converge parameter.

The number in parentheses indicates the degree of change. In actual use, the alternation_converge parameter value is as follows: 30% -> 0.7, 20% -> 0.8, 10% -> 0.9

3. Feature thresholding in optimization

margin_feature_thresh parmater

Using feature weight threholding method (dev in figure), you can get faster computation. ( Shown in time cost page )
Of course, there is some performance loss.