Pooling vs Convolution - Number of Outputs

[tom...@gmx.de]

Hi,

I was wondering why pooling layers are defined differently compared to convolutions. While convolutions round down, pooling layers round up. I find that is an inconsistency which makes many thinks more complicated (e.g. FCN).

Pooling:

pooled_height_ = static_cast<int>(ceil(static_cast<float>(height_ + 2 * pad_h_ - kernel_h_) / stride_h_)) + 1;

Convolution:

const int output_dim = (input_dim + 2 * pad_data[i] - kernel_extent) / stride_data[i] + 1;

My propose is replacing the ceil with floor!

Cheers,
Tom

[Evan Shelhamer]

See https://github.com/BVLC/caffe/issues/1318 for discussion.

Evan Shelhamer

--
You received this message because you are subscribed to the Google Groups "Caffe Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to caffe-users+unsubscribe@googlegroups.com.
To post to this group, send email to caffe...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/caffe-users/05fde386-4fd7-4e67-977f-a8bfa9d9e39c%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.