Query about the Data Balancing

[yxliwhu]

When I read the paper there are some describe about the data balancing

For each training volume containing N occupied voxels, we randomly sample 2N empty voxels from occluded regions for training.

Here I have two questions:

1. According to Figure 2, the occupied voxels is the green area and the occluded regions is the blue area?
2. If yes, when we test our image, the segmentation result is based on the random result of occluded regions. That means (Figure 2 as example)some voxel would not in the region of table or chair, what is the meaning of those voxel in the result?

[JamesTonG321]

My understanding is that the number of object categories is K, the number of predicted classes is K+1, the more 1 class is means empty voxel, that is, the voxel which is in the occluded region but not belongs to the table or chair. Do i understand what you mean? @yxliwhu