difference about logistic_gradient?

[louyanyang]

hi, @eric612 , i'm reading region_loss_layer.cpp、yolov3_layer.cpp in your implementation and comparing region_layer.c 、yolo_layer.c in darknet. i find a difference about "logistic_gradient": yolov2,in your implementation and darknet, the way using logistic_gradient is same, and loss includes logistic_gradient , but in yolov3, your implementation puts logistic_gradient in backward , but there isnot logistic_gradient in darknet, is logistic_gradient not necessary? as i known, your implementation in yolov3 is theoretical, loss including logistic_gradient in yolov2 is also effective. but yolov3 in draknet not use logistic_gradient, can you tell me your idea?

[eric612]

The default value of use_logic_gradient_ was "false" , https://github.com/eric612/MobileNet-YOLO/blob/master/src/caffe/layers/yolov3_layer.cpp#L441-#L471

[unclejokerjoker]

Hello,I wonder why you set the default value of use_logic_gradient_ to false?what is difference between the set of this param to false vs. true？thank you !

[unclejokerjoker]

@eric612 Hello,I wonder why you set the default value of use_logic_gradient_ to false?what is difference between the set of this param to false vs. true？thank you !

[eric612]

@unclejokerjoker

Because darknet didn't use logic_gradient , it only use it in old version "yolov2"

[unclejokerjoker]

@eric612 thanks a lot. further more,why is the param object_scale not used when training?you provide this param to set the value but actually it is not used in the code.Could you please explan about it?thanks !