feature pyramids have been proven powerful in image understanding tasks that
require multi-scale features. State-of-the-art methods for multi-scale feature
learning focus on performing feature interactions across space and scales using
neural networks with a fixed topology. In this pap