Anybody tried SPP layer in Caffe ? Ideas for the classification of the non-centered objects?

[Yongzhe YAN]

Hello everyone,

I would like to know that is there anyone has successfully implemented SPP layer in Caffe which could be found here: http://caffe.berkeleyvision.org/doxygen/classcaffe_1_1SPPLayer.html ? When I tried to implement the SPP layer as the last pooling layer, I found that there comes some errors related to RAM. I'm afraid that they are related to the back_propagation process.

I took a look at the original code of the SPP-Net, they made several modifications on the standard Caffe. So I wondered if it is possible to use the SPP-Layer written in standard caffe.

In addition, in order to classify the objects in varying sizes which is not exactly in the center of the image, any other ideas could help? (I've tried squashing the image but the performance is not quite satisfying)

[Mo Yan]

hello ，I also met this problem ,can you tell me how you solved the problem. thanks.

在 2016年6月21日星期二 UTC+8下午3:04:58，Yongzhe YAN写道：