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

difference of yolov4x-mish.cfg between yours and Alexey #179

Closed erikguo closed 2 years ago

erikguo commented 3 years ago

hi, thank you for your great job!

I found there's one difference of yolov4x-mish.cfg between yours and Alexey.

The activation func just before yolo layer is linear in your yolov4x-mish.cfg while the func is logistic in Alexey version. Why do you change the activation func?

And if I use the version from Alexey, it will report 'don't support'.

@ceccocats

mive93 commented 3 years ago

Hi @erikguo We used the cfg file that Alexey gave us when he firstly released yolov4x-mish. So, it's still the original one, while I think he has changed his. I will check on that.

AlexeyAB commented 3 years ago

Previously logistic activation (sigmoid) was used inside [yolo] layer. Currently we use logistic activation (sigmoid) in the previous conv-layer.

mive93 commented 3 years ago

Fixed with the last commit https://github.com/ceccocats/tkDNN/commit/adac8576b0faf515ad3f459b1f50fd16cef6d64d