search

MrAsimZahid / Dataset-Format-Conversions

This repository contain the codes to convert computer vision, deep learning datasets format in between COCO, YOLO, VOC, and xml.
5 stars 2 forks source link

MOT label format inconsistence #4

Open NeoYY opened 2 years ago

NeoYY commented 2 years ago

Hi!

You extract top left coordinates, width, and height through

    x1 = row.iloc[0]['BB_LEFT']
    y1 = row.iloc[0]['BB_TOP']
    x2 = row.iloc[0]['BB_WIDTH']
    y2 = row.iloc[0]['BB_HEIGHT']

However, you consider x2 and y2 as maximum x and maximum y, which does not sound correct to me.

In addition, from the YOLOv5 repo, they state the label as class x_center y_center width height but what you are using is yolo_x, yolo_w, yolo_y, yolo_h Can you explain the above? Thanks!

MrAsimZahid commented 2 years ago

@NeoYY you are right. This is a logical error. Feel free to create a pull request. Thanks.