Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Keypoint Numbering Messed up for COCO Keypoint export #5048

Closed
jitaxis opened this issue Oct 5, 2022 · 13 comments · Fixed by #5122
Closed

Keypoint Numbering Messed up for COCO Keypoint export #5048

jitaxis opened this issue Oct 5, 2022 · 13 comments · Fixed by #5122
Assignees

Comments

@jitaxis
Copy link

jitaxis commented Oct 5, 2022

My actions before raising this issue

Read/searched the docs
Searched past issues

When exporting keypoint annotations using the COCO keypoint annotation and then importing them back into CVAT, the keypoints are in the wrong order. I don't know if this is useful, but the annotation was done using the keypoint skeleton.

Expected Behaviour

The annotations shouldn't change order when exported. It occurs on the export as I open them on another annotator to check.
This does not occur when exporting in CVAT format.

Current Behaviour

The key points are incorrect order. E.g Square is annotated as below.

1--2
|  |
4--3

When importing this back into CVAT the following may occur.

3--2
|  |
1--4

Possible Solution

There doesn't seem to be a pattern in the changes I can see, but I assume this is happening in the conversion to COCO keypoint.

Steps to Reproduce (for bugs)

  1. Annotate images in CVAT using a keypoint skeleton
  2. Export annotations
  3. Import annotations onto a copy of the dataset to visualise it.

Context

I need this for a project, but this is also a bug in the performance.

Your Environment

I am using the web browser version of CVAT on a linux ubunutu 20.04 OS, with Chrome as my browser.

@bsekachev
Copy link
Member

Hi, @jitaxis

If you are using app.cvat.ai, please share a task/job ID with us.

@jitaxis
Copy link
Author

jitaxis commented Oct 6, 2022

Hi, @jitaxis

If you are using app.cvat.ai, please share a task/job ID with us.

The task ID is:
#15200
The Job ID is:
#15642
Let me know if there is anything else required.

@bsekachev
Copy link
Member

@jitaxis

Looks like I was able to reproduce it.
Is it something like this?

From:
image

To:
image

@bsekachev
Copy link
Member

@yasakova-anastasia Could you please look at the problem?

@jitaxis Thank you for the report.
You can use CVAT for Images format as a temporary workaround.

@yasakova-anastasia yasakova-anastasia self-assigned this Oct 7, 2022
@jitaxis
Copy link
Author

jitaxis commented Oct 9, 2022

@jitaxis

Looks like I was able to reproduce it. Is it something like this?

Yes, that is exactly it.
I need to have my annotations in COCO format for other work, so I won't be able to use CVAT for Images as a workaround for that. Do you know of a conversion script that I could use or how long this issue may take to be fixed?

@SpecLad
Copy link
Contributor

SpecLad commented Oct 10, 2022

I was able to reproduce this on app.cvat.ai, but not on a development server (with the latest commit). It's possible this has already been fixed.

@bsekachev
Copy link
Member

@SpecLad

I am not sure about it.
Try to restore a task from a backup on your local instance.

@jitaxis
Copy link
Author

jitaxis commented Oct 19, 2022

Is the fix able to be used in the docker image of cvat?

@bsekachev
Copy link
Member

@jitaxis

Just pull the changes from the repository and:

image

@bsekachev
Copy link
Member

Probably --build argument is missed.

@jitaxis
Copy link
Author

jitaxis commented Oct 20, 2022

@jitaxis

Just pull the changes from the repository and:

image

I'm still getting the same issue with the key points as before. I pulled the latest "develop" branch and ran the line you mentioned above.
My cvat information is below:
Server version: 2.3
Core version: 7.0.1
Canvas version: 2.15.4
UI version: 1.42.2

Is there something I am missing?

@bsekachev
Copy link
Member

@SpecLad Could you please have a look?

@SpecLad
Copy link
Contributor

SpecLad commented Oct 21, 2022

@jitaxis Did you use the --build option to docker-compose? At the time you posted your comment, the latest UI version should've been higher than 1.42.2, so you might have used outdated Docker images.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging a pull request may close this issue.

4 participants