What and where is the grid size of YOLOv4 detector in Matlab?

As we probably know, the YOLO detector uses {S*S} grid and returns one detection of each anchor boxes from one grid.

However, I cannot find the setting of the grid size S anywhere in the Matlab document below

https://www.mathworks.com/help/vision/ug/object-detection-using-yolov4-deep-learning.html

or anyother documents regarding the YOLO detector.

What is the default setting of the grid size and where to change it?

Best regards,

Answer 1

[Sourav Bairagya]

Hi,

In YOLOv4 object detector, the grid sizes depend on the detector input size. Hence, to change the grid size, you need to change the detector input size accordingly.

For example, default input size in 'csp-darknet53-coco' pretrained model is [608,608,3] and for this three default grid sizes are 19x19, 38x38 and 76x76. If the input size of this model is changed to [416,416,3], then based on the network architecure the grid sizes will be automatically adjusted and new set of grid sizes will be 13x13, 26x26 and 52x52.

To change the detector input size, you can use the 'InputSize' name-value pair argument.

https://www.mathworks.com/help/vision/ref/yolov4objectdetector.html#mw_0e561b1b-ca03-4f11-897e-78f2253f89f4