Particularly poor results are obtained when using the model to infer unlabeled data.

[huanghuang113]

Hello Sir, I am using your model for training my plant point cloud dataset, I modified my dataset based on STPLS3D dataset which has only two semantic classes of stems and culms, and I am getting very good results AP_25 even up to 0.98 during the training process and I am exporting the predicted results of the instances in the validation phase and I am getting satisfactory results. However, when I make inference with unlabeled dataset, the results I get don't seem to be satisfactory, do you know what is the reason for this? @ngoductuanlhp Thank you very much for your great contribution to the 3D point cloud instance segmentation!