Bo Yang

@lapetite123, simply append the sem and ins labels to each point {x, y, z, r, g, b, sem, ins} and then divide the point cloud.

Hi @zhuyijie88, thanks for your interests in our paper. Here's a more detailed discussion about the gradient propagation of Hungarian assignment. Hope it's helpful for your research. https://github.com/Yang7879/3D-BoNet/blob/master/figs/fig_gradient_Hungarian.png

Hi @519830100, the score on val split is few points lower than the final score on test set. This is sensible as the val set has 300 scenes while the...

Hi @BigDeviltjj, the sparse conv is used to predict semantics independently, instead of being integrated into our code. Here are easy steps to do the experiments. https://github.com/Yang7879/3D-BoNet/issues/6#issuecomment-521896317

hi @BigDeviltjj, this is a fundamental question regarding the proposed pipeline. When looking at the raw 3D point clouds (with or without colors), we humans can easily identify clusters/subsets of...

@zhixinwang @519830100 @96lives We don't plan to release the code and model for ScanNet as it relies on the third-party SparseConv whose BSD License is not compatible with the MIT...

hi @thangvubk, basically, it means if the new instance does not have more than 7 pts overlapped with previous instances, that new instance will be treated as a real new...

hi @zz-3280, the instance label is simply starting from 0(or 1 or whichever number you like) and then adding 1 by 1. Basically, that label is the index for an...

hi @turboxin, exactly, there's no need to remove additional bboxes, because each object is uniquely detected and segmented.

Hi @clare19997, please uncomment the lines here for visualization. https://github.com/Yang7879/3D-BoNet/blob/201ae46ffc6a00d5a9fae44da486d3702a83a1a3/main_eval.py#L212