Padding in fully convolutional networks

[Tiferet Gazit]

I am confused about the large padding (pad=100) used at the bottom of fully convolutional networks such as these. Why is this padding needed, rather than simply making sure each layer has a padding that is the correct size relative to its kernel? Also, how do I determine the size to give this large padding?

Thank you!

Tiferet

[Li Sun]

I am also confuse about this. On the final layer, they crop the result back to the size of data. Why not choose a smaller padding size? Anyone can answer?

[Evan Shelhamer]

​From new fcn.berkeleyvision.org FAQ​:

Why pad the input?: The 100 pixel input padding guarantees that the network output can be aligned to the input for any input size in the given datasets, for instance PASCAL VOC. The alignment is handled automatically by net specification and the crop layer. It is possible, though less convenient, to calculate the exact offsets necessary and do away with this amount of padding.

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/e1f6cb68-b627-424d-973a-9c3f63703214%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[klo...@vicomtech.org]

Hello all,


I am also having problems with this padding. How would one calculate the offsets to remove such a large padding?

Thank you very much,

El martes, 27 de septiembre de 2016, 22:05:51 (UTC+2), Evan Shelhamer escribió:

​From new fcn.berkeleyvision.org FAQ​:

Why pad the input?: The 100 pixel input padding guarantees that the network output can be aligned to the input for any input size in the given datasets, for instance PASCAL VOC. The alignment is handled automatically by net specification and the crop layer. It is possible, though less convenient, to calculate the exact offsets necessary and do away with this amount of padding.

Evan Shelhamer

On Sat, Jul 2, 2016 at 2:26 PM, Li Sun <lisu...@gmail.com> wrote:

I am also confuse about this. On the final layer, they crop the result back to the size of data. Why not choose a smaller padding size? Anyone can answer?

On Sunday, June 19, 2016 at 3:02:31 PM UTC+1, Tiferet Gazit wrote:

I am confused about the large padding (pad=100) used at the bottom of fully convolutional networks such as these. Why is this padding needed, rather than simply making sure each layer has a padding that is the correct size relative to its kernel? Also, how do I determine the size to give this large padding?

Thank you!

Tiferet

--
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...@googlegroups.com.

To post to this group, send email to caffe...@googlegroups.com.