YOLO format conversation

[saikrishnadas]

I have a few annotations that is originally in .csv format. I would need to convert it to yolo darknet format inorder to train my model with yolov4. my .cvs file :

YOLO format is : object-class x y width height

where, object_class, widht, height is know from my .csv format. But finding x,y is confusing .Note that x and y are center of rectangle (are not top-left corner).

@AlexeyAB

[stephanecharette]

The normalized label format is described on this page, along with some examples: https://www.ccoderun.ca/programming/2019-08-18_Darknet_training_images/ (Scroll down towards the bottom.)

[saikrishnadas]

@stephanecharette my question is, how will we get the center_x and center_y from the x_min,x_max,y_min,y_max? The document explains the conversion of center_x / the image pixel

[stephanecharette]

If the center is at (10, 10) and the width is 4, height is 4, then from the center to the left is 10 - (4/2), and from the center the right is 10 + (4/2), right? Same with top and bottom.

I strongly suggest you go back and re-read that page carefully. It explains how to get from X and Y coordinates to the normalized values. So if you reverse the process, you also can go from normalized values back to X and Y coordinates.

[saikrishnadas]

@stephanecharette I clearly understand what you saying but i dont have center points. I have only the top_right,bottom_right,top_left_bottom_left which is x_min,x_max,y_min,y_max in my .csv file

[stephanecharette]

You're kidding, right?

[stephanecharette]

If you have all those, then obviously you also have the center points. Add another column to your csv file, and do the math to find the center point. x1 + x2 / 2, same with y.

[saikrishnadas]

Sure

[stephanecharette]

Did a little graph to help visualize the problem: (the center is at 4,4)

So once normalized, you should have:

• x = center_x / image_width = 4 / 10 = 0.4
• y = center_y / image_height = 4 / 10 = 0.4
• w = object_w / image_width = 5 / 10 = 0.5
• h = object_h / image_height = 5 / 10 = 0.5

[saikrishnadas]

@stephanecharette I really appreciate your effort. My image size is 1020 x 720p so, w = obj_w/1020 and h = obj_h/720 right ? or shld it be w = obj_w/480 and h = obj_h/480 as im going to train it on 480x480 yolov4 ?

[lsd1994]

w = obj_w/1020 and h = obj_h/720

[saikrishnadas]

Can't open label file. (This can be normal only if you use MSCOCO): data/obj/Avacado_retraining__172.txt

I get this error

@AlexeyAB @stephanecharette