DCNv2 icon indicating copy to clipboard operation
DCNv2 copied to clipboard

Published 20 hours ago • verified publisher icon CharlesShang

training speed is slow

Open Hub-Tian opened this issue 4 years ago • 1 comments

When I change traditional conv with deformable conv, the training speed is 10✖ times lower than before with a vgg backbone. my pytorch version is v1.1 and my cuda is v10.1. Anyone knows how to improve the speed?

Hub-Tian avatar Apr 04 '20 13:04 Hub-Tian

Hi,

Did you still suffer from low speed?

hua-wu-que avatar Aug 16 '20 00:08 hua-wu-que