Handling rotated bounding boxes when exporting from Label Studio to COCO or YOLO

sajarin · May 13, 2024, 6:50pm

Question:

How can I maintain the rotated bounding box orientation when I export my annotated data from Label Studio into COCO or YOLO formats? The bounding boxes are always exported as non-rotated boxes.

Answer:

The YOLO and COCO dataset formats typically only support axis-aligned bounding boxes and do not account for rotation information. As a result, when you export rotated bounding boxes from Label Studio into these formats, they are converted back to axis-aligned boxes without rotation.

To preserve the rotation information, you must use a format that supports it, such as the Label Studio JSON format. If using YOLO, there is YOLO OBB option in the Label Studio export. LS doesn’t support COCO for rotated bboxes.

Dharini_Karthikeyan · December 21, 2024, 5:45am

Any idea on how to display them using python through the json because when I displayed it the bounding box appears but not correctly over the tilted object. Because of the rotated bounding box I guess.

makseq · December 21, 2024, 12:21pm

Use this:

github.com

HumanSignal/label-studio-sdk/blob/master/src/label_studio_sdk/converter/utils.py#L426-L478


      
          def convert_annotation_to_yolo_obb(label, normalize=True):
              """
              Convert LS annotation to Yolo OBB format.
          
              Args:
                  label (dict): Dictionary containing annotation information including:
                      - original_width (int): Original width of the image.
                      - original_height (int): Original height of the image.
                      - x (float): X-coordinate of the top-left corner of the object in percentage of the original width.
                      - y (float): Y-coordinate of the top-left corner of the object in percentage of the original height.
                      - width (float): Width of the object in percentage of the original width.
                      - height (float): Height of the object in percentage of the original height.
                      - rotation (float, optional): Rotation angle of the object in degrees (default is 0).
                  normalize (bool, optional): Whether to normalize the coordinates to the range [0, 1] (default is True).
          
              Returns:
                  list of tuple or None: List of tuples containing the coordinates of the object in Yolo OBB format.
                      Each tuple represents a corner of the bounding box in the order:
                      (top-left, top-right, bottom-right, bottom-left).
              """

This file has been truncated. show original

Topic		Replies	Views
Image Rotation COCO Export for Object Detection Label Studio Support data-export , annotations	7	83	July 29, 2024
How to export segmentation in COCO/YOLO format? Label Studio Support data-export , yolo	1	1352	May 15, 2024
Err22 for COCO/YOLO export Label Studio Support data-export	3	209	May 23, 2024
Json formats exported using Object Detection with Bounding Boxes include x,y,width,height,rotation. How do you of the remaining three points? Label Studio Support	4	24	November 28, 2024
Export in YOLO format after SAM annotation Label Studio Support data-export , annotations	1	16	December 12, 2024

Handling rotated bounding boxes when exporting from Label Studio to COCO or YOLO

Question:

Answer:

Related topics