bo-zhang-cs / gaic-pytorch Goto Github PK

The paper says that multi-scale features are just concatenated but in your code it's
cat_feat = torch.cat((f3,f4,0.5*f5),1).
Why is f5 feature divided in half prior the concatenation here? Were the shared pretrained models trained with this multiplication?

When I perform"sudo bash make_all.sh",
E=rod_align_api -D_GLIBCXX_USE_CXX11_ABI=0 -gencode=arch=compute_61,code=sm_61 -std=c++14 FAILED: /home/zm/wd/GAIC/untils/rod_align/build/temp.linux-x86_64-3.6/src/rod_align_kernel.o
'Compiling rod_align kernels by nvcc...
In file included from rod_align_kernel.cu:2:0:
rod_align_kernel.h:4:21: fatal error: THC/THC.h: No such file or directory
#include <THC/THC.h>
^
compilation terminated.'
Can you help me with this problem? Looking forward to your reply!