Why the contrast features in the Contrast Block share the same batch normalization layer? - Githubissues

Mhaiyang / ICCV2019_MirrorNet

Code for the ICCV 2019 paper "Where Is My Mirror?"

Other

56 stars 16 forks source link

Why the contrast features in the Contrast Block share the same batch normalization layer? #4

Open steermomo opened 4 years ago

steermomo commented 4 years ago

In the contrast block, why use a shared BN layer instead of using a separate BN layer for each ccl feature?

https://github.com/Mhaiyang/ICCV2019_MirrorNet/blob/032bc60dbd257a35871d68773ceaef9cfd445e05/mirrornet.py#L187 https://github.com/Mhaiyang/ICCV2019_MirrorNet/blob/032bc60dbd257a35871d68773ceaef9cfd445e05/mirrornet.py#L196 https://github.com/Mhaiyang/ICCV2019_MirrorNet/blob/032bc60dbd257a35871d68773ceaef9cfd445e05/mirrornet.py#L202 https://github.com/Mhaiyang/ICCV2019_MirrorNet/blob/032bc60dbd257a35871d68773ceaef9cfd445e05/mirrornet.py#L208 https://github.com/Mhaiyang/ICCV2019_MirrorNet/blob/032bc60dbd257a35871d68773ceaef9cfd445e05/mirrornet.py#L214

lxtGH commented 4 years ago

I have the same question why you set the bn shared with different layers? @Mhaiyang