"tf.batch_gather" is a built-in function matching "tf.nn.top_k" in my own opinion.

In other words, my opinion is maybe a mistake.

The origin "tf.batch_gather" converts local batch_indices into global ones.

• Here is the array example

• The example that the local indices have been converted to global ones.

(The TopKV2 results from "tf.nn.top_k")

So we can find that 0 to 4 and 2 to 6.

then it gathers values corresponding to the indices.

• my batch_gather function

I do not change the batch_indices. In contrast, the data array is processed, and what I do is just a demo to show the function's principle.

please read my codesome_test.py to find more, and I also provide batch_gather.py which is from tensorflow-github.

Wanna know more about batch_gather.py, you can refer https://github.com/tensorflow/tensorflow/blob/r1.13/tensorflow/python/ops/array_ops.py

• PS: the top-k function can be used to hard negative mining in SSD or YOLO, some object detection algorithms I am studying. And the "tf.batch_gather" gathers corresponding yHat array to calculate the loss.