FCN output is shifted relatively to input image

[Ilya Zhenin]

It makes correct prediction but output is a bit shifted relatively to input image. I cant understand why. 

I just feed network with image(not neccesarily square) and get shifted output but after crop with same dimensions.

Did someone ecnountered with such a thing? 

[Evan Shelhamer]

Did you define the crop manually or let coordinate mapping do it for you? Manual crops are susceptible to accidental shifts.

See

https://github.com/shelhamer/fcn.berkeleyvision.org/blob/master/voc-fcn32s/net.py#L3

https://github.com/shelhamer/fcn.berkeleyvision.org/blob/master/voc-fcn32s/net.py#L62

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/12c7daa7-77db-4f35-bbdf-bf24216f2254%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ilya Zhenin]

Fistly yes I did it manually. But wondering if crop layer works otherwise I set it up in deploy(cropping at the end final output layer with original dimensions of data) and output is the same. 

I use caffe version araund april 2016, if it's somehow matters.

среда, 23 ноября 2016 г., 23:34:48 UTC+3 пользователь Evan Shelhamer написал:

Did you define the crop manually or let coordinate mapping do it for you? Manual crops are susceptible to accidental shifts.

See

https://github.com/shelhamer/fcn.berkeleyvision.org/blob/master/voc-fcn32s/net.py#L3

https://github.com/shelhamer/fcn.berkeleyvision.org/blob/master/voc-fcn32s/net.py#L62

Evan Shelhamer

On Wed, Nov 23, 2016 at 2:56 AM, Ilya Zhenin <inf.s...@gmail.com> wrote:

It makes correct prediction but output is a bit shifted relatively to input image. I cant understand why. 

I just feed network with image(not neccesarily square) and get shifted output but after crop with same dimensions.

Did someone ecnountered with such a thing? 

--
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.