search

chenyilun95 / tf-cpn

Cascaded Pyramid Network for Multi-Person Pose Estimation (CVPR 2018)
MIT License
793 stars 197 forks source link

About BN location #10

Closed kaleidoscopical closed 6 years ago

kaleidoscopical commented 6 years ago

Hi! Thank you for making this project open to the world! I run into some confuses because I am new to Tensorflow. I wonder whether you add Batch Normalization(BN) to your added layers, e.g. 1*1 conv kernel and bottleneck in RefineNet. Could you please give me a hint about where you add BN?

chenyilun95 commented 6 years ago

That may be part due to historical problem. When we reproduce FPN for detection, we find BN can stablize the training process in case of NaN loss, thus we inherit from that. In fact that incurs redundant overhead of computation if now it trains well. BN is applied in original bottleneck blocks in resnet. As far as I know, it can improve the performance.

kaleidoscopical commented 6 years ago

Thanks for your quick response.