Actual learning for visual object detection

Actual learning for visual object detection

 

 

This paper is about making training data. Easy normal data is not useful for training so thr effective data should be close to the boundary of false and positive values.

 

The suggested method is like following.

1. It operates interative process. In other words, in the initial step, the generated data may be wrong and not close to the boundary. But as process priceeds, it get close to the boundary value.

2. Human unterferes in that process. So, the later step can refer to human's comments.

 

The paper'a exoeriment suggests that there are ten steps. And in the last step, it could generate many valuable and boundary closing data.

 

So, it said finally as the human labor works get reduced, the resource performance can be improved.