Regarding Input shape of data into CNN deep learning extension
varunm1
New Altair Community Member
Hello
In the current deep learning extension, how is the input shape of CNN considered? In tensorflow, when I train images(converted to pixels) the shape of an array is (nb_samples, rows, columns, channels) for a 2d Conv. How will this happen in CNN of RM? Can we specify the samples? Is there a different convolution 1D or 2D or 3D option that can be chosen which I didn't find in the operator.
@hughesfleming68 any input on this?
Thanks,
Varun
In the current deep learning extension, how is the input shape of CNN considered? In tensorflow, when I train images(converted to pixels) the shape of an array is (nb_samples, rows, columns, channels) for a 2d Conv. How will this happen in CNN of RM? Can we specify the samples? Is there a different convolution 1D or 2D or 3D option that can be chosen which I didn't find in the operator.
@hughesfleming68 any input on this?
Thanks,
Varun
0
Answers
-
Hi Varun, unfortunately I don't have an answer. Perhaps @pschlunder could explain. I have only just started to scratch the surface with some CNN tests on time series problems with DL4J. If anything, I should be asking you questions not the other way around.
Have you taken a look at the DL4J documentation and cheat sheets? https://deeplearning4j.org/docs/latest/deeplearning4j-cheat-sheet#layers-conv
1 -
Hi @hughesfleming68
I have gone through the document, thanks for sharing. It says that normal convolution operator is a 2D, but I am not sure how it's taking (nb_samples,channels, image_rows, image_columns) values. Will try to check it out.
Thanks,
Varun0 -
@pschlunder can you provide some insights on this? Thanks0
