SPP_Net with Fully convolutional networks

[Saeed Izadi]

Hi guys, 

I want to put a SPP_net on top of a fully convolutional network pre-trained on semantic segmentation to extract fixed-size features for bounding boxes with varying sizes. How can I do so? Has SPP_Net been implemented in Caffe? If yes, how can I use it to put it on top a fully convolutional network (before classification layer)? 

Thanks

[Etienne Perot]

Hello, unfortunately, i think SPP layer implemented in Caffe does not keep spatial organization, but you can probably reshape it...