about labeliu

[breadxiexie]

Hello, the fast anomaly detail annotation achieved using Labeliu on the website is very impressive. Can you roughly explain how it was implemented?

[LongHZ140516]

Hi @breadxiexie

Thank you very much for your question. Our anomaly detail annotations use the image and video functions in LabelU.

• For images, we used the global text description and 2D bounding box features (including text descriptions for the specified regions) for annotation. Some feature introductions can be found at https://opendatalab.github.io/labelU/#/guide/task-annotation/image.

• For videos, we used the global text description, segment division (including text descriptions for the segments), and timestamp features for annotation. Detailed feature introductions can also be found at https://opendatalab.github.io/labelU/#/guide/task-annotation/video.

If you are interested in our specific operation process, you may want to check out our project page (https://opendatalab.github.io/LOKI/). We have created dynamic videos in the corresponding content sections for 3D, image, and video annotation processes to help everyone understand our annotation workflow.

[Cwyxx]

In image annotation tasks, is it standard for human-labeled bounding boxes to be represented as [x_min, y_min, width, height]?

[Cwyxx]

@LongHZ140516

[LongHZ140516]

In image annotation tasks, is it standard for human-labeled bounding boxes to be represented as [x_min, y_min, width, height]?在图像标注任务中，人工标注的边界框是否标准地表示为[x_min, y_min, width, height]？

Thank you for your question.

You're absolutely right. Manually labeled bounding boxes typically consist of four parameters: [x, y, width, height], where:

• x and y represent the coordinates of the top-left corner of the bounding box;

• width and height represent the size (width and height) of the box.

For a detailed explanation of the bounding box format, you can refer to the official documentation here: LabelU Bounding Box Schema.

If you have further questions about using the annotation tools, feel free to try the interactive demo here: LabelU Image Annotation Playground. It might help clarify any confusion through hands-on experience.

[Cwyxx]

Thank you very much!!!