search

VinAIResearch / ISBNet

ISBNet: a 3D Point Cloud Instance Segmentation Network with Instance-aware Sampling and Box-aware Dynamic Convolution (CVPR 2023)
Apache License 2.0
115 stars 23 forks source link

Hidden Test Set #46

Closed trandangtrungduc closed 10 months ago

trandangtrungduc commented 10 months ago

My question is "Is the scale of confidence score important when you check the results on benchmark website?" If I have 2 pred_masks with confident scores of 0.9 and 0.6, will the evaluation results be different or similar from if I have those 2 pred_masks with confident scores of 0.4 and 0.1 on a different scale?

Thank you very much.

ngoductuanlhp commented 10 months ago

I have tested on the validation set of ScanNet: regardless of the scale used (0.0 to 1.0 or 0.0 to 2.0), when the relative confidence scores of the predictions remain the same, the results remain consistent.