Clarification Needed on Bounding Box Coordinate Conversion for AI-TOD Dataset in YOLO Format about ai-tod

[Chasel-Tsui]

Hi, we follow the coco format in AI-TOD, the (x, y) in (x, y, w, h) seem to be the top left corner point coordinates. The boxes with negative values may be introduced during the image cropping or labelling process, I think they only account for a very small proportion of the dataset, you can simply filter out them. In order to obtain the YOLO format, you need to convert the coordinates to center coordinates, you need also to normalize them by the image size.