stn loc net could not be trained

dnkirill / stn_idsia_convnet

An implementation of a convolutional neural network with a spatial transformer

67 stars 25 forks source link

Open chengmengli06 opened 6 years ago

chengmengli06 commented 6 years ago

features and weight are all zero, there will be no gradient for both features and weight, both of them will stay to be zero.

userka commented 6 years ago

Hi,

I am running into the same issues. Could the author please comment on that? I would also be interested in the used datasets to reproduce the results.

Thanks a lot.