How is the labeled ratio 0.01%(0.0001) reached?

[fanyan0411]

Thanks for your great work!

I run the "prepare_data_otoc.py", and all the settings are default.
Then, I want to compute label ratio of the whole points,

np.where(sem_labels>-100)[0].shape[0] / len(sem_labels))

and I got about 0.02~0.2.

Besides, in some of .ply files, len(sem_labels) equals about 80,000, so if we want 0.01% label raito, labled points should be 8.

It looks impossible in the "one thing one click" setting. Otherwise, it must be my misunderstanding.

Could you tell me does the computation of labeled ratio I showed is right?

Thank you again for your reply!

[liuzhengzhe]

Thanks for your message.

I re-calculated it on train+val set, and found there are 48737 objects to annotate and there are in total 223890677 points. So it is around 0.02%.

For 0.01% annotations, we randomly sample half of the objects to evaluate the performance of fewer annotations.

Zhengzhe