[core] Add tf.image.flipLeftRight

[annxingyuan]

TF: https://www.tensorflow.org/api_docs/python/tf/image/flip_left_right

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

---

This change is 

[lina128]

Reviewable status: complete! 1 of 1 approvals obtained (waiting on @annxingyuan, @lina128, and @pyu10055)

---

tfjs-backend-cpu/src/kernels/FlipLeftRight.ts, line 40 at r1 (raw file):

      for (let row = 0; row < imageHeight; row++) {
        const rowOffset = row * (imageWidth * numChannels);

Hi Ann, just curious, can we use indexToLoc and locToIndex? I think they can make this block of logic more readable.

[lina128]

Reviewable status: complete! 1 of 1 approvals obtained (waiting on @annxingyuan and @pyu10055)

---

tfjs-backend-cpu/src/kernels/FlipLeftRight.ts, line 40 at r1 (raw file):

Previously, lina128 (Na Li) wrote…

Hi Ann, just curious, can we use indexToLoc and locToIndex? I think they can make this block of logic more readable.

Discussed offline, what I was imagining was something like in this code: https://github.com/tensorflow/tfjs/blob/master/tfjs-backend-cpu/src/kernels/Dilation2D.ts#L61
Essentially, make nestedArray of output and do computation on that, at the end convert that nestedArray to typedArray and write to backend. This is more readable, but I realize there can be performance impact of the two array conversion. So nvm, compute index directly on flattened array sounds good. Also it will be consistent with wasm code.