Support batch input for median_filter2d

[WindQAQ]

Addresses #230. Also meets the interface of mean_filter2d.

A general random test case is down below. Maybe there are some ways to speed up. It's now much more slower than scipy's implementation even for the input of shape [1, h, w, 1].
https://colab.research.google.com/drive/1UkNXxjTMHifKmHB2N5ojOHFojtXrOUIG

[facaiy]

Good job!

[facaiy]

@Mainak431 Hi, would you mind taking a look?

[Mainak431]

@WindQAQ Thanks for the excellent contribution. Everything looks pretty fine.
Can you please add a small training example using this layer with a conv2d layer and dropout layer in tensorflow with maybe mnist dataset? So, that it can be directly used in the network itself?

[WindQAQ]

@Mainak431 Here is the example colab notebook. I will include this in addon's example in maybe the next PR.

[Mainak431]

Nice contribution.

…

On Sun 16 Jun, 2019, 10:20 PM Tzu-Wei Sung, ***@***.***> wrote: @Mainak431 <https://github.com/Mainak431> Here is the example colab notebook <https://colab.research.google.com/drive/1ZOAFTRdxKYDBpG75ZITukIr8MjUR2qDA>. I will include this in addon's example in maybe the next PR. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#288?email_source=notifications&email_token=AIRDMQVU3QCXZFC6XXLN4STP2ZVPDA5CNFSM4HYLJF6KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXZQ3MI#issuecomment-502468017>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AIRDMQT2RQ7ED3MIX6IELALP2ZVPDANCNFSM4HYLJF6A> .

[Squadrick]

@WindQAQ Thanks for the contribution.