| ... | ... | @@ -5,4 +5,6 @@ Make sure to follow these rules: |
|
|
|
- There might be multiple labels in the same images. In this case multiple bounding boxes with multiple class labels need to be drawn.
|
|
|
|
- For labelling images, we want to use the [LabelImg ](https://github.com/HumanSignal/labelImg)tool. (On Windows we had issues when using labelImg with the latest python release 3.12.3 --\> downgrading to a lower version such as 3.9.4 resolved the issues). Make sure to export your labels in the **YOLO** **format**.
|
|
|
|
- To make sure that we have consistent bounding boxes, we define them as follows: "Smallest possible bounding box that still covers the entire label (all details of the label need to be inside of bounding box)". You can also have a look at the [LabelDictionary ](https://gitlab.kit.edu/ukona/aiss_cv_group_6/-/wikis/Labeling-Convention/LabelDictionary)to see some examples for bounding boxes.
|
|
|
|
- To ensure consistency for class names, each label needs to be identified uniquely with its corresponding class name in the [LabelDictionary](https://gitlab.kit.edu/ukona/aiss_cv_group_6/-/wikis/Labeling-Convention/LabelDictionary). Please make sure to check this before labelling images and labels that are not yet included in the label dicitonary. |
|
|
\ No newline at end of file |
|
|
|
- To ensure consistency for class names, each label needs to be identified uniquely with its corresponding class name in the [LabelDictionary](https://gitlab.kit.edu/ukona/aiss_cv_group_6/-/wikis/Labeling-Convention/LabelDictionary). Please make sure to check this before labelling images and labels that are not yet included in the label dicitonary.
|
|
|
|
|
|
|
|
When naming a new label class, try to give a highly descriptive name (e.g. bio_hexagon_label instead of bio_1) |
|
|
\ No newline at end of file |
