search

VinAIResearch / ISBNet

ISBNet: a 3D Point Cloud Instance Segmentation Network with Instance-aware Sampling and Box-aware Dynamic Convolution (CVPR 2023)
BSD 3-Clause "New" or "Revised" License
123 stars 23 forks source link

How to test with STPLS3D no.26-28 datasets without labels? #17

Closed Sophiscaty-C closed 1 year ago

Sophiscaty-C commented 1 year ago

I try to test with STPLS3D no.26-28 datasets (by changing config prefix to 'test') but I got this error: image Thanks! Looking forward to replies.

ngoductuanlhp commented 1 year ago

This is a bug in the load function of isbnet/data/stpls3d.py. I fixed it in the latest commit. You could check it and try again.

Thank you for your report.

Sophiscaty-C commented 1 year ago

Thanks!That worked. And seems like the result labels are all added with an extra 1 which making the range come to 2-15, is that also a bug? image

ngoductuanlhp commented 1 year ago

For the STPLS3D dataset, there are 14 foreground classes, then the predicted class_id ranges from 0->13. During evaluation, we add with an extra 1 to distinguish them from the background class (0). Thus the cls value ranges from 1->14.

Sophiscaty-C commented 1 year ago

Thanks for your replies!

huanghuang113 commented 1 year ago

Hi@Sophiscaty-C, how did you test the stpls3d dataset, do you need to change ''prefix: 'val250m' '' to, ''prefix: 'test' '', I put the test file in the test folder, using python3 tools/test.py configs/isbnet stpls3d.yaml work_dirs/stpls3d/exp/latest.pth Can I use the model for testing with this command?

WS517 commented 5 months ago

@Sophiscaty-C,how did you test the stpls3d dataset,Can you elaborate on the procedure, thank you very much!