s9xie / DSN

Deeply-supervised Nets
http://vcl.ucsd.edu/~sxie/2014/09/12/dsn-project/
Other
141 stars 66 forks source link

Number of parameters seems larger than NIN paper unlike your paper statement? #9

Open erogol opened 8 years ago

erogol commented 8 years ago

It seems like you use the same skeleton with the NIN model bu the supervised layers. And any number is same like channel numbers and filter sizes. Even you introduces more params with supervised layers. however your paper states that number of parameters are arranged to keep the values comparable with the original NIN model.