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
104 stars 22 forks source link

Performance on ScanNetv2-200 datasets and config files #62

Open shjung13 opened 2 weeks ago

shjung13 commented 2 weeks ago

Hello authors,

I was running the code on the ScanNetv2-200 dataset and found the config is not matching with its training counterparts.

It can't load some parameters in instance mask branch and outputs 0 mAP.

After I change the num_dyco_layer from 3 to 2, I was able to reproduce the performance, so the default parameter has to be 2 not 3.

Would this be the right case?

Thank you.