Training problem about superpoint_graph

[loicland]

hi,

seems like you ran out of memory on the second point cloud of testred, which is weird.

a few questions:

• did you use the suggested parametrization to compute the partition? Can you check with visualize that the partitions of the reduced graphs went well?
• did you uncomment the line 129 #print('Ptn with', clouds.size(0), 'clouds') of learning/pointnet?
• do you have other processes running on your card? A 1080Ti should be amply sufficient
• what version of pytorch are you using?

[dlliyang]

thanks very much.my pytorch is 0.4.0,and when i change it to 0.2.0,it's ok now.thank you !

[loicland]

yes that's what I thought, 0.4.0 breaks the memory monger. I will fix it soon and let you know when it's done. Thanks for pointing it out.

it should work with 0.3.0 as well.