[Feature] Add support for the focal Tversky loss #2783

zifuwanggg · 2023-03-22T20:37:53Z

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

The focal Tversky loss was proposed in https://arxiv.org/abs/1810.07842. It has nearly 600 citations and has been shown to be extremely useful for highly imbalanced (medical) datasets. To add support for the focal Tversky loss, only few lines of changes are needed for the Tversky loss.

Modification

Add gamma as (optional) argument in the constructor of TverskyLoss. This parameter is then passed to tversky_loss to compute the focal Tversky loss.

BC-breaking (Optional)

Does the modification introduce changes that break the backward-compatibility of the downstream repos?
If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.

Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases here, and update the documentation.

Checklist

Pre-commit or other linting tools are used to fix the potential lint issues.
The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDet or MMDet3D.
The documentation has been modified accordingly, like docstring or example tutorials.

CLAassistant · 2023-03-22T20:37:58Z

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
1 out of 2 committers have signed the CLA.

✅ zifuwanggg
❌ Zifu Wang

Zifu Wang seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

zifuwanggg · 2023-03-22T21:27:54Z

It seems that I have signed CLA but still get complains.

codecov · 2023-03-23T03:42:52Z

Codecov Report

Patch coverage: 33.33% and project coverage change: -0.02 ⚠️

Comparison is base (d6079bc) 88.36% compared to head (af32881) 88.34%.

❗ Current head af32881 differs from pull request most recent head b618e06. Consider uploading reports for the commit b618e06 to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #2783      +/-   ##
==========================================
- Coverage   88.36%   88.34%   -0.02%     
==========================================
  Files         149      149              
  Lines        9109     9112       +3     
  Branches     1523     1524       +1     
==========================================
+ Hits         8049     8050       +1     
- Misses        810      811       +1     
- Partials      250      251       +1

Flag	Coverage Δ
unittests	`88.34% <33.33%> (-0.02%)`	⬇️

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

Impacted Files	Coverage Δ
mmseg/models/losses/tversky_loss.py	`91.22% <33.33%> (-3.22%)`	⬇️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

xiexinch · 2023-03-23T06:46:51Z

Hi @zifuwanggg,
Thanks for your contribution!
You can check if the email used in the local commit is the same as the email in the GitHub profile, if not, please add the email used in the commit to the GitHub profile.

MengzhangLI · 2023-03-23T07:39:16Z

mmseg/models/losses/tversky_loss.py

@@ -75,6 +78,9 @@ class TverskyLoss(nn.Module):
        beta (float, in [0, 1]):
            The coefficient of false negatives. Default: 0.7.
            Note: alpha + beta = 1.
+        gamma (float, in [1, 3]): The focal term. When `gamma` > 1, the loss


We may add assertion below to ensure 1.0 < gamma < 3.0, so does unit test.

The [1, 3] range was mentioned in the paper, although they do not mention why gamma should be in this range. Theoretically, gamma can be any real number. Practically, a small gamma (e.g. < 2) works better. mmseg.models.losses.focal_loss does not specify the range of gamma, so perhaps we should remove the requirement for gamma in [1, 3] and let the user to decide?

Beside, I also think alpha and beta in the original TverskyLoss class does not need to satisfy assert alpha + beta == 1.0. Floating point equality is dangerous and according the definition of the Tversky index, the only requirement for alpha and beta is that they are non-negative numbers.

xiexinch · 2023-03-23T07:48:09Z

mmseg/models/losses/tversky_loss.py

+            if gamma > 1:
+                tversky_loss **= (1 / gamma)


Might add a unit test.

zifuwanggg · 2023-03-23T17:35:57Z

Hi, @xiexinch

Thanks for the comment. The email I used for the very first commits was indeed by default username@localhost.local. Even if I add that email to my GitHub account, I cannot verify that email address since it cannot receive the verification email from GitHub. Modify the email address for previous commits are strongly discouraged. Is there a workaround to let me sign the CLA, or I might just close this one and open a new pull request?

xiexinch · 2023-03-24T03:28:02Z

Hi, @xiexinch

Thanks for the comment. The email I used for the very first commits was indeed by default username@localhost.local. Even if I add that email to my GitHub account, I cannot verify that email address since it cannot receive the verification email from GitHub. Modify the email address for previous commits are strongly discouraged. Is there a workaround to let me sign the CLA, or I might just close this one and open a new pull request?

Hi @zifuwanggg,
In that case, I think it's better to close this pull request and create a new one :)

OpenMMLab-Assistant-007 · 2023-04-13T07:16:15Z

Hi
@zifuwanggg
We are grateful for your efforts in helping improve this open-source project during your personal time.
Welcome to join OpenMMLab Special Interest Group (SIG) private channel on Discord, where you can share your experiences, ideas, and build connections with like-minded peers. To join the SIG channel, simply message moderator— OpenMMLab on Discord or briefly share your open-source contributions in the #introductions channel and we will assist you. Look forward to seeing you there! Join us ：https://discord.gg/UjgXkPWNqA
If you have a WeChat account，welcome to join our community on WeChat. You can add our assistant ：openmmlabwx. Please add "mmsig + Github ID" as a remark when adding friends：）
Thank you again for your contribution❤

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers. ## Motivation The focal Tversky loss was proposed in https://arxiv.org/abs/1810.07842. It has nearly 600 citations and has been shown to be extremely useful for highly imbalanced (medical) datasets. To add support for the focal Tversky loss, only few lines of changes are needed for the Tversky loss. ## Modification Add `gamma` as (optional) argument in the constructor of `TverskyLoss`. This parameter is then passed to `tversky_loss` to compute the focal Tversky loss. ## BC-breaking (Optional) Does the modification introduce changes that break the backward-compatibility of the downstream repos? If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR. ## Use cases (Optional) If this PR introduces a new feature, it is better to list some use cases here, and update the documentation. ## Checklist 1. Pre-commit or other linting tools are used to fix the potential lint issues. 2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness. 3. If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDet or MMDet3D. 4. The documentation has been modified accordingly, like docstring or example tutorials. Reopening of previous [PR](#2783).

Add support for the focal Tversky loss

f8c67f8

mm-assistant bot assigned MeowZheng Mar 22, 2023

Zifu Wang added 2 commits March 22, 2023 21:45

Add support for the focal Tversky loss

9b723ce

Add support for the focal Tversky loss

af32881

xiexinch changed the title ~~Add support for the focal Tversky loss~~ [Feature] Add support for the focal Tversky loss Mar 23, 2023

MengzhangLI reviewed Mar 23, 2023

View reviewed changes

xiexinch reviewed Mar 23, 2023

View reviewed changes

zifuwanggg added 6 commits March 23, 2023 13:28

Add support for the focal Tversky loss

025280a

Add assertion for gamma

8e7fb21

Add assertion for gamma

315471a

Add assertion for gamma

5486cde

Add assertion for gamma

bed7188

Add assertion for gamma

b618e06

zifuwanggg closed this Mar 24, 2023

zifuwanggg deleted the the-focal-tversky-loss branch March 24, 2023 09:30

zifuwanggg mentioned this pull request Mar 24, 2023

[Feature] Add support for the focal Tversky loss #2791

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Feature] Add support for the focal Tversky loss #2783

[Feature] Add support for the focal Tversky loss #2783

zifuwanggg commented Mar 22, 2023

CLAassistant commented Mar 22, 2023 •

edited

Loading

zifuwanggg commented Mar 22, 2023

codecov bot commented Mar 23, 2023 •

edited

Loading

xiexinch commented Mar 23, 2023

MengzhangLI Mar 23, 2023 •

edited

Loading

zifuwanggg Mar 23, 2023 •

edited

Loading

zifuwanggg Mar 23, 2023

xiexinch Mar 23, 2023

zifuwanggg commented Mar 23, 2023 •

edited

Loading

xiexinch commented Mar 24, 2023

OpenMMLab-Assistant-007 commented Apr 13, 2023

[Feature] Add support for the focal Tversky loss #2783

[Feature] Add support for the focal Tversky loss #2783

Conversation

zifuwanggg commented Mar 22, 2023

Motivation

Modification

BC-breaking (Optional)

Use cases (Optional)

Checklist

CLAassistant commented Mar 22, 2023 • edited Loading

zifuwanggg commented Mar 22, 2023

codecov bot commented Mar 23, 2023 • edited Loading

Codecov Report

xiexinch commented Mar 23, 2023

MengzhangLI Mar 23, 2023 • edited Loading

Choose a reason for hiding this comment

zifuwanggg Mar 23, 2023 • edited Loading

Choose a reason for hiding this comment

zifuwanggg Mar 23, 2023

Choose a reason for hiding this comment

xiexinch Mar 23, 2023

Choose a reason for hiding this comment

zifuwanggg commented Mar 23, 2023 • edited Loading

xiexinch commented Mar 24, 2023

OpenMMLab-Assistant-007 commented Apr 13, 2023

CLAassistant commented Mar 22, 2023 •

edited

Loading

codecov bot commented Mar 23, 2023 •

edited

Loading

MengzhangLI Mar 23, 2023 •

edited

Loading

zifuwanggg Mar 23, 2023 •

edited

Loading

zifuwanggg commented Mar 23, 2023 •

edited

Loading