Incorrect bounding boxes

[Thomaspatrick14]

Hi,

I am trying to obtain the bounding boxes from an image that i feed the model. The yolov5 model from torch hub has been loaded on to cuda. I get the right bounding boxes when i feed the image like this :

but when i load the image to cuda, for some reason yolo expects an image with dimensions a multiple of 32 (ig has to do something with the kernel size), so i try it like this:

but this way the bounding boxes are incorrect and there are a lot of detections.