Regarding  loss、precision、recall issue encountered in training superpoints

Hello author, after loading the sp_v6 model you provided, I conducted feature point detection and then trained it again, and found that the best results is loss 2.9573, precision 0.4417, recall 0.7610. May I ask when you  train the sp_v6 model  What are the final results of loss,precision, recall these three values? Thank you very much