I have a Yolov8-n detection model that simply detects two classes with bounding boxes and confidences: cat and dog. When I visualize the model with Netron, the last array shows as 1x6x10710 as shown below. The raw prediction numpy array from the model is saved as a .bin file that I read:

array = np.fromfile(yolo_data_path)

The size of the .bin file is exacly 257,040. Since the output is float32 [1,6,10710], so the output size seems to be correct: 4x6x10710 = 257,040

enter image description here

Now given that I have this .bin output array, how can I read it and decode the bbox and print the final detections? Basically, the bounding boxes, class and their confidences. I assume that 6 is basically 4 bounding boxes, confidence and class number.

Looking at the Yolov8 predict.py, I understand somehow after parsing these preditions, we have to pass them through non_max_suppression() to get the final results.