Running time 4 times slower than SENet

[xwang-icsi]

Thank you for your re-implementation of GENet, when I was trying to run GENet with Theta-minus, the parameter free version, it seems like the running time will be about 4 times slower than the SENet counterpart, I found that F.interpolate is the main reason for slowing down the running process, not sure whether you can get close running time?

[ma-xu]

pytorch/pytorch#12409 (comment)

[ma-xu]

Thank you for your re-implementation of GENet, when I was trying to run GENet with Theta-minus, the parameter free version, it seems like the running time will be about 4 times slower than the SENet counterpart, I found that F.interpolate is the main reason for slowing down the running process, not sure whether you can get close running time?

It's east to solve this problem, u just need to update your pytorch 1.1 to pytorch 1.2.0. The issue is addressed in 1.2 version.