Some instances that lean together are recognized as the same

[huanghuang113]

Hello author, thank you very much for your great contribution on the top of 3D point cloud segmentation, when I am using your model, maybe there is a problem with the parameter setting of my knn algorithm, so when the model is speculated, some two instances which are leaning on one piece will be recognized as the same instance, I would like to know how to solve this kind of problem? You can see in the image below that the two arrowed leaves are recognized as one instance.

Is this case a problem with the knn parameter settings and how did you solve this problem?

[yyds12]

@huanghuang113 hello, I would like to use the model as you do, for example, I made my dataset in S3DIS format, but still need an oversegmentation file, how should this file be generated?