How to convert YOLOv8 raw output to bounding box coordinates and class probabilities

[Santabot123]

I trained a custom object detection model using ultralytics and converted it to Tensorflow Lite format.

Now when I make a prediction for an image (640x640x3) using the tflite model, the result is a tensor of the shape [1, 7, 8400].

I found out that in [1, 7, 8400]:

1 - is batch size

7 - 4 bounding box coordinates(x_center, y_center, width, height) + 3 probability each class

8400 - 640 pixels/8 =80; 80x80=6400. 640 pixels/16=40; 40x40= 1600. 640 pixels/32=20; 20x20=400. 6400+1600+400=8400

I am looking for a way to decode this tensor to bounding box coordinates and class probabilities. Is there any ready-made solution ?