googlenet: Resize the network's input image size

[201power]

Hi all,

I tried to use caffe for deep dream. 

I trained a googlenet using single channel gray-scale image (instead of 3 channels). 

I tried to resize the network's input image size for deep dreaming, however, I got an error:

src=net.blobs['data']

src.reshape(1,1,h,w)

h,w is the new input image size

Error:

Check failed: K_ == new_K (2048 vs 3200) Input size incompatible with inner product parameters.

However, for 3 channel image

src.reshape(1,3,h,w) worked fine. 

Any suggestions? Thanks!

[201power]

The layer I used for dream is inception_4c/output