search

prs-eth / PanopticSegForLargeScalePointCloud

Code and dataset of paper "TOWARDS ACCURATE INSTANCE SEGMENTATION IN LARGE-SCALE LIDAR POINT CLOUDS"
MIT License
73 stars 9 forks source link

How do inference? #2

Closed cama36 closed 1 year ago

cama36 commented 1 year ago

Hello, thank you very much for your work. I noticed that when using eval.py, the data will also load semantic labels and instance labels. I think this part is only for evaluation, but when the data does not have this content, I modify this part of the code, and I will get surprising results. How can I do inference?

bxiang233 commented 1 year ago

Hello, thank you very much for your work. I noticed that when using eval.py, the data will also load semantic labels and instance labels. I think this part is only for evaluation, but when the data does not have this content, I modify this part of the code, and I will get surprising results. How can I do inference?

Hi, thanks for your interest in our work! Yes, the semantic labels and instance labels are for evaluation only. I think the easiest way to do the quick inference is by adding "fake labels". For example, all points assign instance labels with 0, and semantic labels with 0. Then you will not get reasonable value for evaluation metrics, but you could still get prediction and visualization output files. Hope it can solve your problem.

cama36 commented 1 year ago

Thank you very much for your reply,I see.