Training VN_DGCNN_partseg on own dataset about vnn HOT 2 CLOSED

Alright, so here's the story in short:

• I fixed it.
• I got the solution right away, weeks ago, but I adjusted the wrong model (vn_pointnet) by putting the number of classes (1 in this case) in the conv7 layer. So naturally nothing happened at that point. Derp.
• This was an unnecessary post, but hopefully still useful for amateurs like myself.

from vnn.

Hi there!
I haved trained the VN_DGCNN model on my own dataset with 4096 point_num per PointCloud,while it took up a lot of video memory.When I set the batchsize to 22 with 4096 pointnum per frame, it almost took up totally 60G video memory on three GeForce RTX 3090(with 24G per card).
So I would like to ask if you also have the problem of using too much memory during your training.
Thank you!

from vnn.