About the "instance_sizes"

[lin1061991611]

Hello,i noticed that there is a (6,3) shape array named "instance_sizes" in train.py，what does it mean? please

[LiuShihHung]

It means the average instance sizes in the whole training data(height, width, length).

[lin1061991611]

It means the average instance sizes in the whole training data(height, width, length).

Thanks for your reply, and I noticed that there is a list named "sem_r" in main_eval.py, what does it mean?

[KiKuBoy]

为什么取预测实例的大小?

[ammaradel]

How to calculate the average instance sizes and the semantic radius values? And why are the instance_sizes of length 6?

[meidachen]

From my understanding, these instance sizes are used as "ground truth" for Input Size Prediction in figure 3 of the original paper. Basically, it will try to map a center from the center prediction network to one of the defined instance sizes. And later, using this instance size for pointnet to extract the local features.

The issue I had with this is that I'm trying to segment outdoor scenes and would like to get the instance segmentation for buildings and cars. And these two types of objects are very different in size.

If someone has a better understanding of this, I would be happy to learn. Or if the authors would like to respond to this issue...