Recognition of Point Sets Objects in Realistic Scenes

Figure 11

The entire training process. (a) The accuracy of the training process (which is on the training set) as a function of the number of iterations. It appears that the accuracy of the training set increases with the number of iterations. When the accuracy finally reaches 98.8%, it can be said that the training process has completely learned the characteristics of various point cloud data. (b) The value of the cross-entropy loss. The smaller the value, the smaller the predicted deviation with the actual deviation, the better the prediction by the model, cross-entropy loss value decreases as the number of iterations increases. (c) The change of the learning rate during the learning rate training process. The learning rate is attenuated with a certain decay rate as the number of iterations increases. (d) A graph that records the decay of the learning rate as the number of iterations increases during training. As the number of iterations increases, the rate of decay of the learning rate increases.

(a)

(b)

(c)

(d)