-
-
Notifications
You must be signed in to change notification settings - Fork 886
Description
Search before asking
- I have searched the X-AnyLabeling issues and found no similar feature requests.
Description
Background
X-AnyLabeling supports the difficult attribute in annotation shapes, which is very useful when annotators are uncertain about the correct class of an object and want to mark it for later expert review.
However, at present, there is no way in the GUI to:
Filter images by difficult=true
Review only difficult objects
Efficiently re-annotate and clear the difficult flag after expert confirmation
This makes the difficult attribute hard to use in real production pipelines.
Use case
In real annotation projects, many samples cannot be confidently labeled by non-expert annotators. They are marked as difficult=true and need to be reviewed later by experts.
Currently, users must write scripts to extract those samples into a new folder, which breaks the original dataset structure and increases maintenance cost.
Built-in difficult filtering would significantly improve productivity and data quality.
Benefit
Enables closed-loop annotation workflow
Makes difficult a truly useful metadata field
Reduces dependence on external scripts
Improves annotation quality control
Additional
Add a Difficult Filter / Review Mode in the GUI.
Example design:
Menu:
View → Filter Objects → Difficult = True
Behavior:
Only show images that contain at least one shape with "difficult": true
Optionally hide non-difficult shapes in the same image
Allow annotators to:
Modify labels
Uncheck difficult
Save and go to next difficult sample
This would enable a complete expert-review workflow inside X-AnyLabeling without external scripts.
Are you willing to submit a PR?
- Yes I'd like to help by submitting a PR!