search

ayooshkathuria / YOLO_v3_tutorial_from_scratch

Accompanying code for Paperspace tutorial series "How to Implement YOLO v3 Object Detector from Scratch"
https://blog.paperspace.com/how-to-implement-a-yolo-object-detector-in-pytorch/
2.32k stars 724 forks source link

Why does batch norm layer has the parameter of weight and bias? #65

Open AsakusaRinne opened 4 years ago

AsakusaRinne commented 4 years ago

In this tutorial, when explaining the weight file, it mentioned that if a convolutional layer has batch norm, then it will get 4 parameters from the weight file. The 4 parameters are: bn_weight, bn_bias, bn_running_mean, bn_running_var. But for batch normalization, why do we still need bn_weight and bn_bias? What kind of role do the bn_weight and bn_bias cast? I am a freshman of DL so my question may seem to be dull...I will appreciate it if you could tell me the answer or where to find the answer.

bot66 commented 4 years ago

just google batch normal definition, batch normal having learnable parameters , not just simply normalize the input data, every epoch will update its parameters.

AsakusaRinne commented 4 years ago

just google batch normal definition, batch normal having learnable parameters , not just simply normalize the input data, every epoch will update its parameters.

Thank you very much!