added Grad-CAM implementation

[ruthcfong]

Here's an implementation of Grad-CAM that extends SaliencyMask class.

I've included a few optional arguments in the GetMask function (i.e., should_resize, three_dims) to allow the option to return the original, low-res 2D Grad-CAM heatmap, which I need.

[googlebot]

Thanks for your pull request. It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

📝 Please visit https://cla.developers.google.com/ to sign.

Once you've signed, please reply here (e.g. I signed it!) and we'll verify. Thanks.

---

• If you've already signed a CLA, it's possible we don't have your GitHub username or you're using a different email address. Check your existing CLA data and verify that your email is set on your git commits.
• If your company signed a CLA, they designated a Point of Contact who decides which employees are authorized to participate. You may need to contact the Point of Contact for your company and ask to be added to the group of authorized contributors. If you don't know who your Point of Contact is, direct the project maintainer to go/cla#troubleshoot.
• In order to pass this check, please resolve this problem and have the pull request author add another comment and the bot will run again.

[ruthcfong]

I signed it!

[googlebot]

CLAs look good, thanks!

[dsmilkov]

Thanks. This is a great contribution! One comment regarding dependency on skimage.

[dsmilkov]

Since the core library is trying to avoid dependencies, how about using tf.image.resize_bilinear(...) instead of skimage

[dsmilkov]

One more comment: Update README.md bullet list by adding Grad-CAM to the list

[ruthcfong]

I've replaced skimage with tf's resize function as suggested and updated the README file.

[dsmilkov]

Awesome. Thanks for this great contribution!

[nsthorat]

+1 thanks for the contribution! :)

[nalinimsingh]

@ruthcfong -- I'm curious about this line in the GradCAM implementation here.

Why is this matrix initialized with ones -- from my read of the paper, it seems it should be filled with zeroes?

[This doesn't affect the qualitative trend you see across the image in the GradCAM masks I've generated, since this is a constant shift at each pixel, but it does slightly perturb some of the numeric values.]

[ruthcfong]

@nalinimsingh unfortunately, I don't remember what my reasoning was when I added the GradCAM implementation. I think you're right though that the proper way to do it is to initialize with zeros.

My hypothesis that I probably was looking at this keras implementation by jacobgil (committed March 2017; my commit was July 2017), which does the same thing.

I created a new pull request with the fix.