Implemented GIoU by briankosw · Pull Request #347 · Lightning-Universe/lightning-bolts

briankosw · 2020-11-09T10:28:43Z

What does this PR do?

Implements Generalized Intersection over Union as mentioned in #251

Before submitting

Was this discussed/approved via a Github issue? (no need for typos and docs improvements)
Did you read the contributor guideline, Pull Request section?
Did you make sure your PR does only one thing, instead of bundling different changes together? Otherwise, we ask you to create a separate PR for every change.
Did you make sure to update the documentation with your changes?
Did you write any new necessary tests?
Did you verify new and existing tests pass locally with your changes?
If you made a notable change (that affects users), did you update the CHANGELOG?

PR review

Is this pull request ready for review? (if not, please submit in draft mode)

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

codecov · 2020-11-09T10:30:37Z

Codecov Report

Merging #347 (1e0b2d1) into master (4309cab) will increase coverage by 0.07%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #347      +/-   ##
==========================================
+ Coverage   80.44%   80.52%   +0.07%     
==========================================
  Files         101      103       +2     
  Lines        5687     5710      +23     
==========================================
+ Hits         4575     4598      +23     
  Misses       1112     1112

Flag	Coverage Δ
cpu	`25.36% <100.00%> (+0.30%)`	⬆️
pytest	`25.36% <100.00%> (+0.30%)`	⬆️
unittests	`80.08% <100.00%> (+0.08%)`	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
pl_bolts/losses/object_detection.py	`100.00% <100.00%> (ø)`
pl_bolts/metrics/object_detection.py	`100.00% <100.00%> (ø)`

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4309cab...1e0b2d1. Read the comment docs.

akihironitta · 2020-11-15T00:13:30Z

@briankosw Thank you for your contribution! Let us know if you need help :]

briankosw · 2020-11-15T09:11:50Z

Hi @akihironitta! I'm trying to write tests for GIoU, but I don't have experience with unittest (I'm more used to pytest) and it seems that the majority of the tests in this repo weren't fully setup. I'll submit a commit soon, and could you take a look at review it for me?

akihironitta · 2020-11-15T09:37:50Z

@briankosw Sure! Mark this PR as ready for review when you're ready :]

jeffydc

Great to see GIoU loss in bolts!
Can you add the docs?

akihironitta

Thank you for the swift action! Would you mind having a look?

akihironitta

LGTM!

briankosw · 2020-11-16T02:32:25Z

I'm going to add a little bit more to the docstring and fix that isort failure, but that should be the final touches!

akihironitta · 2020-11-17T11:59:11Z

@briankosw Also, please remember to add the summary of your change to the changelog.

Borda

pls follow @akihironitta suggestions

annikabrundyn · 2020-11-18T00:24:41Z

This is awesome! Does it make sense to also add this to the PL metrics?

Borda · 2020-11-18T10:01:21Z

This is awesome! Does it make sense to also add this to the PL metrics?

I would keep it here for a little while and later like any other metric we may move them to PL, we just need to try they are stable and kind of standard :]

ananyahjha93 · 2020-11-18T16:38:43Z

@annikabrundyn @briankosw @Borda let's add this to PL as a class metric + the current functional form.

briankosw · 2020-11-19T02:06:08Z

I'd be happy to close this PR and open a new one in PL as soon as I'm given the go ahead 🙂. If so, I'm assuming #370 and part of #251 should be migrated over as well?

Borda · 2020-11-19T07:20:55Z

I'd be happy to close this PR and open a new one in PL as soon as I'm given the go ahead 🙂. If so, I'm assuming #370 and part of #251 should be migrated over as well?

hi, the point is that we shall merge it here, and later (let's say a few weeks) we add it in PL and eventually drop from here

briankosw · 2020-12-02T13:08:52Z

I think this PR is ready for a final review. Would love your reviews @Borda @ydcjeff.

jeffydc · 2020-12-04T07:27:41Z

Thank you for your work @briankosw
Is there any difference between this and this one from torchvision except that we extract giou from 1 in this?

briankosw · 2020-12-04T10:38:08Z

Thank you for your work @briankosw
Is there any difference between this and this one from torchvision except that we extract giou from 1 in this?

Ah I didn't realize torchvision has its own implementation. You're correct in that we subtract giou from 1. Should I change it to use torchvision's giou?

jeffydc · 2020-12-11T06:53:47Z

Thank you for your work @briankosw
Is there any difference between this and this one from torchvision except that we extract giou from 1 in this?

Ah I didn't realize torchvision has its own implementation. You're correct in that we subtract giou from 1. Should I change it to use torchvision's giou?

I think both are right. in giou loss, we subtract giou from 1. Since torchvision has giou, we only need to 1 - giou to get giou loss... So it can be said this PR is unneeded, Thank you for your work btw.

But let's see core members' thoughts too!

akihironitta · 2020-12-12T21:34:02Z

@briankosw @ydcjeff
Thank you for your work! In my opinion, it's still nice to have the loss here or in PL as the current function and also a subclass of pl.metrics.Metric like pl.metrics.Accuracy as @ananyahjha93 suggested.

@annikabrundyn @briankosw @Borda let's add this to PL as a class metric + the current functional form.

I think it's reasonable to have the loss in Bolts and later migrate it to PL like @Borda mentioned.

@Borda @ananyahjha93 So, how shall we move on?

Borda · 2020-12-13T09:19:14Z

yes, as we do not have any loss in PL, we shall also think if it is reasonable to include it there... eventually we would move the losses as a sub-package... so lets it land here and later talk about losses as a complete functional block :]

Co-authored-by: Jeff Yang <ydcjeff@outlook.com>

Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>

pep8speaks · 2020-12-20T22:51:59Z

Hello @briankosw! Thanks for updating this PR.

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2020-12-20 23:19:48 UTC

briankosw marked this pull request as ready for review November 15, 2020 10:45

jeffydc reviewed Nov 15, 2020

View reviewed changes

Comment thread pl_bolts/losses/giou.py Outdated

jeffydc reviewed Nov 15, 2020

View reviewed changes

Comment thread tests/losses/test_giou.py Outdated

jeffydc reviewed Nov 15, 2020

View reviewed changes

akihironitta suggested changes Nov 16, 2020

View reviewed changes

Comment thread pl_bolts/losses/giou.py Outdated

akihironitta approved these changes Nov 16, 2020

View reviewed changes

briankosw requested a review from jeffydc November 16, 2020 04:01

briankosw mentioned this pull request Nov 16, 2020

Add Intersection over Union loss #370

Closed

Borda added the enhancement New feature or request label Nov 16, 2020

Borda reviewed Nov 16, 2020

View reviewed changes

Comment thread pl_bolts/losses/giou.py Outdated

Comment thread tests/losses/test_giou.py Outdated

briankosw requested a review from Borda November 17, 2020 11:44

Borda reviewed Nov 17, 2020

View reviewed changes

Comment thread pl_bolts/losses/giou.py Outdated

briankosw requested a review from Borda November 18, 2020 10:14

briankosw and others added 20 commits December 20, 2020 23:47

Fixing error

b80828c

isort test

d289d9d

Renaming file to object_detection.py

e0a68c4

Reflecting module name change in test

7f7bec9

Implemented GIoU

8ebd93b

Changing max/min to torch.max/torch.min

b4ac0e1

Adding tests for GIoU

b32b369

Updated docstring

b5e5d85

Co-authored-by: Jeff Yang <ydcjeff@outlook.com>

Fixed isort error and added link to paper in docstring

2e29634

Parametrizing tests using pytest

d61ed9b

Adding changelog and removing eps

b6a4dbb

Fixing error

a42e111

isort test

8370947

Renaming file to object_detection.py

721e734

Reflecting module name change in test

edffe9f

Update pl_bolts/losses/object_detection.py

f1d1646

Co-authored-by: Jirka Borovec <Borda@users.noreply.github.com>

Updating docstring for giou

637792b

Fixing code formatting

34c1987

Adding doctest

730bcdc

Adding tests for giou

8fa52f7

refactor

ebda4b0

Borda approved these changes Dec 20, 2020

View reviewed changes

Borda added 2 commits December 21, 2020 00:09

fix

d684d2b

format

1e0b2d1

Borda merged commit a72766c into Lightning-Universe:master Dec 20, 2020

briankosw mentioned this pull request Dec 21, 2020

Intersection over Union Metric/Loss #469

Merged

8 tasks

Borda added this to the v0.3 milestone Jan 18, 2021

This was referenced Mar 13, 2021

[tune](deps): Bump pytorch-lightning-bolts from 0.2.5 to 0.3.0 in /python/requirements suquark/ray#8

Closed

[tune](deps): Bump pytorch-lightning-bolts from 0.2.5 to 0.3.0 in /python/requirements sven1977/ray#8

Closed

Conversation

briankosw commented Nov 9, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does this PR do?

Before submitting

PR review

Did you have fun?

Uh oh!

codecov Bot commented Nov 9, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

akihironitta commented Nov 15, 2020

Uh oh!

briankosw commented Nov 15, 2020

Uh oh!

akihironitta commented Nov 15, 2020

Uh oh!

Uh oh!

Uh oh!

jeffydc left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

akihironitta left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

akihironitta left a comment

Choose a reason for hiding this comment

Uh oh!

briankosw commented Nov 16, 2020

Uh oh!

Uh oh!

Uh oh!

akihironitta commented Nov 17, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Borda left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

annikabrundyn commented Nov 18, 2020

Uh oh!

Borda commented Nov 18, 2020

Uh oh!

ananyahjha93 commented Nov 18, 2020

Uh oh!

briankosw commented Nov 19, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Borda commented Nov 19, 2020

Uh oh!

briankosw commented Dec 2, 2020

Uh oh!

jeffydc commented Dec 4, 2020

Uh oh!

briankosw commented Dec 4, 2020

Uh oh!

jeffydc commented Dec 11, 2020

Uh oh!

akihironitta commented Dec 12, 2020

Uh oh!

Borda commented Dec 13, 2020

Uh oh!

pep8speaks commented Dec 20, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Comment last updated at 2020-12-20 23:19:48 UTC

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

briankosw commented Nov 9, 2020 •

edited

Loading

codecov Bot commented Nov 9, 2020 •

edited

Loading

jeffydc left a comment •

edited

Loading

akihironitta commented Nov 17, 2020 •

edited

Loading

briankosw commented Nov 19, 2020 •

edited

Loading

pep8speaks commented Dec 20, 2020 •

edited

Loading