Update README.md

README.md

@@ -92,7 +92,8 @@ For token-level annotation of referring expressions, see the companion dataset [
 Each entry in the dataset contains the following fields:
 
 - `image`: The original image file.  
-- `mask`: A binary segmentation mask isolating the target object.  
+- `mask`: A binary segmentation mask isolating the target object.
+- `related_masks`: A list of other masks of the class in the image, will be empty for a `single_precence` entry.
 - `boxed_image`: The original image overlaid with a red bounding box highlighting the target object.  
 - `box_xmin`, `box_xmax`, `box_ymin`, `box_ymax`: The normalized bounding‑box coordinates.  
 - `is_coco`: A binary flag (1 for COCO-class, 0 for non‑COCO).