Skip to content
This repository has been archived by the owner on Dec 19, 2024. It is now read-only.

Ann ids are not unique when converting to the COCO format #217

Open
Robotatron opened this issue Jul 28, 2022 · 3 comments
Open

Ann ids are not unique when converting to the COCO format #217

Robotatron opened this issue Jul 28, 2022 · 3 comments
Labels
bug Something isn't working

Comments

@Robotatron
Copy link

Describe the Bug:
When using datasetinsights to convert the dataset for instance segmentation to the COCO format, the annotation IDs generated in the COCO are not unique. Frameworks such as Detectron2 will refuse to train because of this.

How to Reproduce?
Using datasetinsights to convert the dataset for instance segmentation to the COCO format

What did you expect to happen:
All annotation IDs are unique

Console logs / stack traces

Screenshots
image

Anything else you would like to add:
Generated several datasets with Unity Perception to confirm. Every generated dataset converted with datasetinsights had duplicate ann ids.

Environment:

  • OS + version: Debian 10
  • datasetinsights version: The latest commit with Instance Segmentation Support - bf6f824
@Robotatron Robotatron added the bug Something isn't working label Jul 28, 2022
@tanker768
Copy link

any follow up on this please?

@Robotatron
Copy link
Author

any follow up on this please?

Do you have the same issue? Would be nice to have someone else to confirm it (although I triple checked it on my own)

@tanker768
Copy link

any follow up on this please?

Do you have the same issue? Would be nice to have someone else to confirm it (although I triple checked it on my own)

yeah same like yours

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
bug Something isn't working
Projects
None yet
Development

No branches or pull requests

2 participants