Data of RL-Trainning

[essentialsa]

there is a picture in the paper which shows the training process of the RL. The input of the RL seems like the Incomplete PointCloud. But in your code I found that you use complete data for all training.Incomplete data just for testing, is that Is this consistent with the paper？