Swapped width and height in DefaultBoxGenerator

[nlgranger]

In the DefaultBoxGenerator module, the image sizes in the image_size argument are given as (height, width) by SSD model but the code assumes (width, height).

vision/torchvision/models/detection/anchor_utils.py

Lines 216 to 217 in 15a9a93

	x_f_k = image_size[0] / self.steps[k]
	y_f_k = image_size[1] / self.steps[k]

I think this has gone unnoticed because the images are usually resized into squares.

cc @datumbox