How to do inference on labelled pointcloud

thangvubk / SoftGroup

[CVPR 2022 Oral] SoftGroup for Instance Segmentation on 3D Point Clouds

MIT License

345 stars 81 forks source link

How to do inference on labelled pointcloud #127

Open Mythili-kannan opened 2 years ago

Mythili-kannan commented 2 years ago

I have trained a custom model using the pretained model by following the mentioned step, now i want to test unlabelled pointcloud(.ply), the dataloader s3dis.py requires labels, to load the data.

can you please help me here.

thangvubk commented 2 years ago

Hi, inference on data without label is supported from commit cf88d9be41ae83a70f9100856a3ca15ee4ddcee9. Please check updated readme for the guideline. You may adapt it to your custom dataset.

Mythili-kannan commented 2 years ago

Thank you so much, i have trained data on custom data set, but 2 or 3 objects are predicted as same instance, i played with scaling and grouping radius parameters still facing same, and the loss also didnt get lower than 0.5, any suggestion please