BoNet for outdoor datasets
vvaibhav08 opened this issue · comments
Hi @Yang7879,
How would you prepare outdoor datasets for BoNet and how do you think BoNet will perform on them? I think we would need to change quite a few things in the pre-processing steps. From the top of my head, increasing the block size from 1mX1m to something bigger would be a must since most outdoor instances are bigger than that.
Hi @vvaibhav08 , if the backbone network is the pnet++ as we used in the paper, I agree that to partition the outdoor point clouds into slightly larger blocks is enough to make it work.
However, pnet++ is unable to deal with a bit larger blocks because of its heavy computation and memory. To achieve satisfactory efficiency and effectiveness on extremely large-scale outdoor point clouds might be a bit challenging.
To better deal with large-scale point clouds, we recently introduced a new backbone network named RandLa-Net to efficiently and effectively process large-scale point clouds in a single pass. Welcome to have a try.
Paper: https://arxiv.org/abs/1911.11236
Project page: https://github.com/QingyongHu/RandLA-Net