Closed ZaiwangLi closed 4 years ago
Please use template
I suspect you might be using spconv-1.1. In that case, you need to add use_hash=False
as an additional argument every time you instantiate a sparse convolution operation (or sub-manifold sparse convolution) in scn.py
. An easier way out is to follow instructions in INSTALL.md
and checkout a specific branch.
@peiyunh Thanks for reply. Hey guys, I haven't try @peiyunh suggestion.
I solved this problem by this answer: [https://github.com/traveller59/spconv/issues/74]
@peiyunh Thanks for reply. Hey guys, I haven't try @peiyunh suggestion.
I solved this problem by this answer: [https://github.com/[traveller59/spconv/issues/74](https://github.com/traveller59/spconv/issues/74)]
I suggest you use my version of spconv.
During training, I meet the problem below at epoch 63 with --nproc_per_node=2, samples_per_gpu=6 and workers_per_gpu=6 at epoch 81 with --nproc_per_node=2, samples_per_gpu=4 and workers_per_gpu=4
cudahash: Completely failed to build Cuda error in file '/root/spconv/src/cuhash/hash_table.cpp' in line 194 : an illegal memory access was encountered.