Report that the grasp poses generated by gen_labels are incorrect.

[lidc54]

As depicted, the generated grasp points are on both sides of the plane; is the collision avoidance not functioning? Moreover, the model trained with this data has not converged, which is likely due to the data itself. However, I've seen reports on CSDN of it running smoothly. Now, I'm unsure how to troubleshoot the issue. Identify where the error lies.

Below is the status of the environmental configuration:

graspnetAPI                    1.2.10
torch                          1.13.0
torchvision                    0.14.0
open3d                         0.10.0.0
numpy                          1.22.0

Here are the downloaded data. ├ ├── scenes │   ├── test_novel.zip │   ├── test_seen.zip │   ├── test_similar.zip │   ├── train_1.zip │   ├── train_2.zip │   ├── train_3.zip │   └── train_4.zip ├── collision_label.zip ├── dex_models.zip ├── rect_labels.zip ├── models.zip └── grasp_label.zip