search

yunshin / SphericalMask

Official implementation of "Spherical Mask: Coarse-to-Fine 3D Point Cloud Instance Segmentation with Spherical Representation"
Apache License 2.0
37 stars 2 forks source link

Requesting config file or pretrain weights. #4

Open frankkim1108 opened 3 months ago

frankkim1108 commented 3 months ago

Hello @yunshin

Thank you for your research and allowing your code to be open sourced.

Currently, I tried to reproduce SphericalMask with the S3DIS dataset. Since SphericalMask is based on ISBNet, I used the config file used in ISBNet to run the S3DIS dataset. However, the SpheircalMask result was much worse than the result shown in the paper.

Could you share the config file or pretrained weights used in this experiment? Thank you for your time.

yunshin commented 3 months ago

Hello, Thank you for your interest. We are going to upload the details of S3DIS accordingly with the conference at the latest.

frankkim1108 commented 3 months ago

Thank you for your response @yunshin !

I have some additional questions about your model.

  1. When you train for Scannet benchmark, do you initialize with ISBNet benchmark pre-train checkpoints?

  2. For training SPHMask benchmark can you kindly share the parameter values of epoch, repeat, learning-rate?

Thank you for your time.