Hi, I was doing profiling on the recently released StarGAN-V2 ( https://github.com/clovaai/stargan-v2 ) pre-trained generator and style_encoder networks. However, certain operations used in these networks are missing from torch profile. Following are the missing handlers:
"aten::leaky_relu"
"aten::to"
"aten::detach"
"aten::floor"
"aten::upsample_nearest2d"
"aten::index"
"aten::to"
"aten::stack"
Please suggest where can I add these operations in profile.py to do an accurate profiling?
The profiling result is correct as all these operations you listed do not require MACs. I've updated the package accordingly and the warning should be fixed now. Thanks!
Hi, I was doing profiling on the recently released StarGAN-V2 ( https://github.com/clovaai/stargan-v2 ) pre-trained generator and style_encoder networks. However, certain operations used in these networks are missing from torch profile. Following are the missing handlers:
Please suggest where can I add these operations in profile.py to do an accurate profiling?