Questions about the 3D match training data

[shinshiner]

Hello, thanks for sharing the well-organized code. I have downloaded the training data of 3dmatch from the link provided in scripts/download_datasets.sh. I found there only 7000+ scene pairs with 0.3 overlap threshold while there are tens of thousands frames in even one sequence of the original 3dmatch dataset. Have you conducted some sampling operations and are these all the training data?

Thanks in advance :)

[chrischoy]

The commercial depth sensors generate noisy samples that are not suitable for registration. A typical pipleline first integrate 50-100 frames into one point cloud. We also followed this procedure and created N^2 pairs within one sequence and then removed pairs < 0.3 overlap.

[shinshiner]

Got it. Thanks for your reply.

[shinshiner]

BTW, it seems that the judgement of overlap is conducted in point clouds with 0.1 voxel size downsampling while 0.05 for the test data, is that true ?