Why don't you use 'batchnorm' in your model?

rongliangzi / Dense-Scale-Network-for-Crowd-Counting

An unofficial implement of paper "Dense Scale Network for Crowd Counting", link: https://arxiv.org/abs/1906.09707

31 stars 8 forks source link

Why don't you use 'batchnorm' in your model? #2

Closed seekFire closed 4 years ago

rongliangzi commented 4 years ago

In experiments on shanghai tech part A dataset, including batch norm does not show improvement. Up to now I have not gotten theoritical explanation for that.

seekFire commented 4 years ago

@rongliangzi Ok, but whether the training process becomes harder without batchnorm?

rongliangzi commented 4 years ago

the loss reduce normally in training process.

seekFire commented 4 years ago

@rongliangzi, OK, Thank you very much!