Evaluate method in test.py

[vdvchen]

Thank you very much for your great work ,but I have some problem when trying to evaluate the results on S3DIS with test.py you provided.
According to the original paper of PointNet(http://openaccess.thecvf.com/content_cvpr_2017/papers/Qi_PointNet_Deep_Learning_CVPR_2017_paper.pdf),when testing on the S3DIS dataset, all points in each block are tested.But in your test.py,I find you sampled 4096 points in each block to test on.Dose it mean your evaluate protocol is different from that in PointNet?
Looking forward to your reply.

[Xinlong Wang]

@vdvchen Hi, we use the same evaluate protocal with PointNet. You can check the official PointNet code. I also asked the PointNet author the same question and decided to follow their setting for fair comparision. I post the reply I got below.

In a strict test setting, we should use all the original points (could be more than 4096 or less than 4096). For convenience, in our scripts we are testing upsample/downsample to a block to a fixed number of 4096 points. The results should be very similar though. I think it is okay to go with the 4096 points.