InceptionV3 (master branch)

[hello-fri-end]

This is a follow up for #2848 which was built on ann-vtable branch and kept on crashing on my system. This PR builds the InceptionV3 layer from the master branch. All the inception modules & corresponding tests have been implemented. The test for all the modules pass except for Inception3B module where the assertion REQUIRE (arma::accu(delta) == 0 ) fails. It seems there's some issue with the backward function of convolution layer but I'm unable to figure it out. Otherwise, this is ready for review @rcurtin @zoq.

PS: This may be helpful for reviewing.
Pytorch Implementation of InceptionV3

[zoq]

You are great, I'll test this out and see if I can reproduce the convolution issue.

[hello-fri-end]

Sounds great, thanks !

[zoq]

I was able to reproduce the issue, but I'm also not quite sure what the problem is, at least not yet.

[hello-fri-end]

I tried debugging the issue by adding print statements in the code and found out at line 287 in convolution_impl.hpp (https://github.com/mlpack/mlpack/blob/master/src/mlpack/methods/ann/layer/convolution_impl.hpp#L287, when outputis added to gTemp.slice(inMap + batchCount * inSize) , some garbage values are added ( maybe because of shape mismatch, not sure).

output.print() before line 287:

    0        0        0        0        0        0
    0        0        0        0        0        0
    0        0        0        0        0        0
    0        0        0        0        0        0
    0        0        0        0        0        0
    0        0        0        0        0        0

gTemp.slice(inMap + batchCount * inSize).print() before line 287:

    0        0        0        0        0
    0        0        0        0        0
    0        0        0        0        0
    0        0        0        0        0
    0        0        0        0        0

gTemp.slice(inMap + batchCount * inSize).print() after line 287:

        0            0            0            0            0  2.6512e-316
        0            0            0            0  6.3734e-322  4.9407e-324
        0            0            0            0            0  4.9407e-322
        0            0            0            0  2.6538e-316  1.5169e-154
        0            0            0            0  2.6383e-316   5.0149e-90
        0            0            0            0  2.6540e-316  6.9886e+228

[mlpack-bot]

This issue has been automatically marked as stale because it has not had any recent activity. It will be closed in 7 days if no further activity occurs. Thank you for your contributions! 👍

[hello-fri-end]

Keep open, please.