EncodeBase64 and DecodeBase64 ops

[vabarbosa]

the TF implementation of the pix2pix model which fails conversion because of

`Unsupported Ops: DecodeJpeg, EncodePng, DecodeBase64`

the Open NSFW model also fails conversion with some of the same ops (tensorflow/tfjs#433).

i wanted to try to implement some of these ops in TensorFlow.js. starting with this pull request for DecodeBase64 and EncodeBase64.

along with this tfjs-core PR, there is a corresponding PR in tfjs-converter (tensorflow/tfjs-converter#376)

---

To see the logs from the Cloud Build CI, please join either
our discussion
or announcement mailing list.

---

This change is 

[googlebot]

Thanks for your pull request. It looks like this may be your first contribution to a Google open source project (if not, look below for help). 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 (or fixed any issues), please reply here (e.g. I signed it!) and we'll verify it.

---

What to do if you already signed the CLA

Individual signers

• It's possible we don't have your GitHub username or you're using a different email address on your commit. Check your existing CLA data and verify that your email is set on your git commits.

Corporate signers

• Your company has a Point of Contact who decides which employees are authorized to participate. Ask your POC to be added to the group of authorized contributors. If you don't know who your Point of Contact is, direct the Google project maintainer to go/cla#troubleshoot (Public version).
• The email used to register you as an authorized contributor must be the email used for the Git commit. Check your existing CLA data and verify that your email is set on your git commits.
• The email used to register you as an authorized contributor must also be attached to your GitHub account.

ℹ️ Googlers: Go here for more info.

[googlebot]

CLAs look good, thanks!

ℹ️ Googlers: Go here for more info.

[nsthorat]

Can you fix the failing errors: https://console.cloud.google.com/gcr/builds/b214fa07-9b38-4194-909d-f5bf95a89811?project=834911136599

Just curious have you tried this in Node, does it work everywhere? Why aren't you using the built in methods to do base64 conversion?

[nsthorat]

By the way thanks for the PR!

[vabarbosa]

Can you fix the failing errors: https://console.cloud.google.com/gcr/builds/b214fa07-9b38-4194-909d-f5bf95a89811?project=834911136599

@nsthorat my apologies, the errors should be resolved now.

Just curious have you tried this in Node, does it work everywhere? Why aren't you using the built in methods to do base64 conversion?

i have added it to the Node kernel (tensorflow/tfjs-node#259) and i tried it out.

regarding built in methods are you asking about btoa()/atob()? if so, browsers fail if a character exceeds the range of a 8-bit byte (0x00~0xFF). for example, this would not work:

btoa('add emphasis— with em dash')

because of the em dash/long dash unicode character (i.e., 0x2014).
if you are referring to some other built in methods please let me and i can review to make sure i didn't miss something.

thanks

[nsthorat]

Hi @vabarbosa!

So @dsmilkov and I just did a deep dive and we have the following conclusions:

• We are actually about to change the internal representation of string tensors to hold onto to the underlying byte array
• This means that for you, you will just need to take that byte array and generate the base64 encoded version (as well as the reverse). This means just the method arrayBufferToString

We will let you know once string stuff is done!

[vabarbosa]

thank you! i'll be on the look out for your string tensor updates. and if you have any questions for me in the meantime, feel free to ask.

[nsthorat]

Here is the PR if you want to follow: #1816

[vabarbosa]

hi @nsthorat i have pulled in all the latest updates (including string tensor PR #1816)
and i have made my changes accordingly. let me know if you have any questions.
thanks.