Why doesnt activations() function work on regression output layer in a CNN imported from caffe using ImportCaffeNetwork?

1 次查看（过去 30 天）

Ahmed Bilal 2020-3-4

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/509025-why-doesnt-activations-function-work-on-regression-output-layer-in-a-cnn-imported-from-caffe-usin

Hi, I have a pretrained Image to Image Regression network in caffe. I have imported it into Matlab (2017b) using importCaffeNetwork(). I am using activations functions in matlab to visualize the output

activations(cnn30,imtest,'conv4','OutputAs', 'channels');

Where 'conv4' is the last layer. So activation of conv4 is my output image. but it gives following error.

-----

Assignment has more non-singleton rhs dimensions than non-singleton subscripts

Error in SeriesNetwork/activations (line 788)

Y(indices{:}) = reshapeFcn(YChannelFormat, numel(observationIndex));

-----

Interesting thing is that it works totally fine on layers above the last layer. for example same activations works fine on 'conv3' and returns a 96x96x8 array.

Plz help.