Add support to NonMaxSuppressionV5

[lina128]

FEATURE Add support to NonMaxSuppressionV5: add arg softNmsSigma and implementation to nonMaxSuppression op.

---

This change is 

[nsthorat]

Great work on this Na! I don't have any major comments, this fits very well in line with our style :)

Just want to double check: you've made sure this matches TensorFlow python by testing in a colab or something of that sort?

Reviewed 8 of 8 files at r1.
Reviewable status: complete! 1 of 1 approvals obtained (waiting on @lina128, @nsthorat, and @pyu10055)

---

tfjs-core/src/backends/array_util.ts, line 3 at r1 (raw file):

/**
 * @license
 * Copyright 2018 Google LLC. All Rights Reserved.

2019

---

tfjs-core/src/backends/array_util_test.ts, line 3 at r1 (raw file):

/**
 * @license
 * Copyright 2018 Google LLC. All Rights Reserved.

2019

---

tfjs-core/src/ops/image_ops.ts, line 148 at r1 (raw file):

/**
 * Performs non maximum suppression of bounding boxes based on
 * iou (intersection over union). This op also supports a Soft-NMS mode (c.f.

nit: put a break line before the new sentence

[lina128]

Thank you for the review Nikhil! I only made sure it matches with TensorFlow c++ code. I will test the python version and let you know. Good catch :)

[googlebot]

All (the pull request submitter and all commit authors) CLAs are signed, but one or more commits were authored or co-authored by someone other than the pull request submitter.

We need to confirm that all authors are ok with their commits being contributed to this project. Please have them confirm that by leaving a comment that contains only @googlebot I consent. in this pull request.

Note to project maintainer: There may be cases where the author cannot leave a comment, or the comment is not properly detected as consent. In those cases, you can manually confirm consent of the commit author(s), and set the cla label to yes (if enabled on your project).

ℹ️ Googlers: Go here for more info.

[pyu10055]

Great work Na! Are there any TF tests for NMSV5 we can include here?

Reviewable status: complete! 2 of 1 approvals obtained (waiting on @lina128 and @pyu10055)

---

tfjs-core/src/ops/image_ops.ts, line 237 at r1 (raw file):

    scoreThreshold = Number.NEGATIVE_INFINITY;
  }
  if (softNmsSigma == null) {

make sure softNmsSigma is between 0 and 1?

[lina128]

Great work Na! Are there any TF tests for NMSV5 we can include here?

Reviewable status: complete! 2 of 1 approvals obtained (waiting on @lina128 and @pyu10055)

tfjs-core/src/ops/image_ops.ts, line 237 at r1 (raw file):

    scoreThreshold = Number.NEGATIVE_INFINITY;
  }
  if (softNmsSigma == null) {

make sure softNmsSigma is between 0 and 1?

Thank you for the review, Ping! Added the check.

[googlebot]

CLAs look good, thanks!

ℹ️ Googlers: Go here for more info.

[lina128]

Great work on this Na! I don't have any major comments, this fits very well in line with our style :)

Just want to double check: you've made sure this matches TensorFlow python by testing in a colab or something of that sort?

Reviewed 8 of 8 files at r1.
Reviewable status: complete! 1 of 1 approvals obtained (waiting on @lina128, @nsthorat, and @pyu10055)

tfjs-core/src/backends/array_util.ts, line 3 at r1 (raw file):

/**
 * @license
 * Copyright 2018 Google LLC. All Rights Reserved.

2019

tfjs-core/src/backends/array_util_test.ts, line 3 at r1 (raw file):

/**
 * @license
 * Copyright 2018 Google LLC. All Rights Reserved.

2019

tfjs-core/src/ops/image_ops.ts, line 148 at r1 (raw file):

/**
 * Performs non maximum suppression of bounding boxes based on
 * iou (intersection over union). This op also supports a Soft-NMS mode (c.f.

nit: put a break line before the new sentence

Done

[lina128]

@nsthorat and @pyu10055 , the python version outputs the same result as the added test in image_ops_test.ts > 'select from three clusters with SoftNMS', here is the Colab link for the Python result: https://colab.research.google.com/drive/1ivLrOjw70UccrM6ENszizE1mb1mGB7RF

[mattyaodoit]

Hi, I am using the latest version of tfjs and still getting this error from my browser

[lina128]

Hi, I am using the latest version of tfjs and still getting this error from my browser

Hi @mattyao1984, this PR implemented the cpu kernel and gpu kernel for NonMaxSuppressionV5 op, and this change is included in tfjs-core 1.5.1 and tfjs 1.5.1. But we haven't included the op in tfjs-converter yet. We will do another release tomorrow to include everything. Stay tuned.

[mattyaodoit]

@lina128 Thanks!