Enhance UI for Export Selection of `Unwanted` Images

[lcyyyao]

Search before asking

• I have searched the X-AnyLabeling issues and found no similar feature requests.

Description

Currently, X-AnyLabeling does not provide a way to explicitly confirm that an image has no objects to annotate. In object detection workflows (e.g., YOLO format), this creates ambiguity: if an image has no bounding boxes, it may be interpreted as either "unlabeled" or "confirmed empty." However, for training purposes, especially in active learning or human-in-the-loop pipelines, it is important to include verified negative samples (images with no targets) as part of the dataset, with proper empty .txt files during export.

Use case

We are developing a garbage pile detection model using drone imagery. During data annotation, some images are visually reviewed and confirmed to contain no garbage pile, and thus should be used as negative samples in training.
X-AnyLabeling does not currently provide a way to mark such images as "confirmed negative," and may exclude them during export or mix them up with unprocessed images. This introduces risk of confusion or data leakage in the training process.

Additional

We suggest adding a feature like:

A "Mark as negative sample" or "Confirm no object" button in the annotation UI

A visual tag or state in the image list (e.g., checkmark, gray icon, etc.)

Proper export behavior: these images should be included in the export with empty .txt (YOLO) or empty annotations (COCO)

This feature would make X-AnyLabeling much more suitable for high-quality iterative annotation workflows. We'd be happy to provide additional feedback or test the feature if needed. Thanks for your great work!

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[CVHub520]

Hello there! 👋 @lcyyyao

Thanks for your detailed suggestion and support! We really appreciate you taking the time to provide such thoughtful feedback.

Regarding the export of empty .txt files for images with no objects, X-AnyLabeling actually supports this by default in the export settings. If you're not seeing this option, please make sure you've updated to the latest version, v3.1.1.

For marking negative samples directly within the annotation UI, you can currently leverage the Flags feature. You could set up two flag options, for instance, positive and negative (with positive as the default). This allows you to explicitly tag images. You can find more details on using flags in our image-level classification example here

We absolutely understand your point about the UI/UX for explicitly marking "no object" images and making this workflow more intuitive. We'll definitely take your suggestions into serious consideration for future updates, exploring how we can integrate this more seamlessly into the UI and interaction flow. Your insights are invaluable!

We encourage you to keep an eye on our progress or even consider submitting a PR if you're inclined to help us build this feature.

Your contributions would be warmly welcomed! 🚀

[lcyyyao]

Thank you for your reply.

Let me clarify my request: after using AI auto-annotation, all images are auto-selected in the lower-right export panel, but I can't uncheck any of them. My goal is to:

Keep some AI-misdetected images (after removing the wrong boxes) and export them as negative samples (empty labels);

Exclude other images that have no ground truth and no AI prediction, so they don’t get exported at all.

Right now, I can’t do this because I can’t deselect any image — it would be helpful if the UI allows me to manually uncheck images/folders from export after AI auto-labeling.

This is important for iterative training workflows, where we only want to include meaningful data (true positives and confirmed negatives), and avoid training on irrelevant or ambiguous images.

Also, it’s possible that this feature already exists, and I just didn’t find it. If so, could you please let me know how to use it?

Thanks again!

[CVHub520]

Ah, thanks so much for clarifying your workflow! That makes perfect sense. 🤔

For your specific need to exclude certain images from export after AI auto-labeling – for instance, those that "terrible" samples that you don't want to include at all – X-AnyLabeling has a dedicated feature for this. You can use the image label deletion function (keyboard shortcut: Ctrl+Shift+Delete). This action will pop out the current image and its associated labels from your active task and move them to an external backup directory. This way, these "unwanted" samples won't be included when you perform your export.

You can find more detailed information on this. We highly recommend taking a few minutes to scan through the document; it often clarifies how to achieve various advanced workflows with X-AnyLabeling.

Hope this helps streamline your iterative training process! Let us know if you have any further questions.

[lcyyyao]

Thank you so much! I’ll take a closer look at the documentation and try out this feature. Really appreciate your help! 🙏