search

ceccocats / tkDNN

Deep neural network library and toolkit to do high performace inference on NVIDIA jetson platforms
GNU General Public License v2.0
718 stars 209 forks source link

Add swish #249

Open AdamCuellar opened 3 years ago

AdamCuellar commented 3 years ago

I've added swish for yolov4-csp-swish and yolov4-csp-x-swish from Darknet

I don't have a Xavier or Nano to test on but for V100:

FP32: 75.49 AVG FPS for yolov4-csp-swish 50.99 AVG FPS for yolov4-csp-x-swish

FP16: 111.95 AVG FPS for yolov4-csp-swish 78.14 AVG FPS for yolov4-csp-x-swish

mive93 commented 2 years ago

Hi @AdamCuellar,

I don't know why I missed this. I will look into it next week, test it on embedded boards, and in case merge it. Thank you.

harsco-jfernandez commented 2 years ago

Hi,

Any reason why the swish implementation from @AdamCuellar hasn't been merged yet? It would be nice to have that available even tho I could replace swish with mish, but there are some advantages to swish.

Thanks.