3DSSD
3DSSD copied to clipboard
Differences between 3DSSD and VoteNet
Hi all, thanks for releasing this great work.
I'm wondering what are the main differences between 3DSSD and VoteNet on the architecture level, besides removing FP layers, fusion sampling strategy in SA layers and 3D Center-ness Assignment Strategy. Since I noticed that VoteNet works poorly on KITTI https://github.com/facebookresearch/votenet/issues/53#issuecomment-584548785, but 3DSSD works just great. I think the two methods share many similarities but why the results are so different.
Looking forward to your insight~