Why is the time usage to precess images with different size differs significantly?

[xujingao13]

I'm just a little confused, as far as I know, all the input images need initially to be resized at the same dimension (1024 * 1024, referred from utils.py) before processed by the framework. So why the time required to process different images with various size differs significantly? For example, I input three different dimensions of the same image (1920 * 1080, 960 * 540, and 480 * 270), the time usage is 1.42s, 0.95s, and 0.57s, respectively. Why is the time usage differs significantly? Don't all images need to be resized at the same dimension initially?