search

ZENGYIMING-EAMON / CorrNet3D

CorrNet3D: Unsupervised End-to-end Learning of Dense Correspondence for 3D Point Clouds
GNU General Public License v3.0
137 stars 21 forks source link

Regarding the pre-trained model performance #7

Closed yanagi-lmd closed 2 years ago

yanagi-lmd commented 3 years ago

I trained the CorrNet3D model following the instruction, however, I could not reproduce the performance of the provided pre-trained model. May I ask you to provide the best seed value or learning curve, if possible?

ZENGYIMING-EAMON commented 2 years ago

Feel free to check the original log record files produced by Pytorch Lightning automatically, and you can check all loss curves by the tensorboard. But since the tensorboard can't be edited, so you may extract the loss curve data from it if you need it (I haven't tried that). Unfortunately, the seed is generated randomly by PyTorch Lightning (see L428) and only printed on the screen, so there is no record. Feel free to ask if you have further questions.