Hybrid Skip: A Biologically Inspired Skip Connection for the UNet Architecture

TL;DR

This paper introduces a biologically inspired hybrid skip connection for UNet that improves segmentation accuracy by balancing edge preservation and texture transfer, but faces challenges in dense regression tasks like depth estimation.

Contribution

The paper proposes HybridSkip connections, a novel skip connection method inspired by perceptual illusions, enhancing UNet's ability to balance detail and smoothness in dense predictions.

Findings

HybridSkip improves segmentation edge preservation.

HybridSkip reduces texture transfer artifacts.

HybridSkip balances detail and smoothness in depth estimation.

Abstract

In this work we introduce a biologically inspired long-range skip connection for the UNet architecture that relies on the perceptual illusion of hybrid images, being images that simultaneously encode two images. The fusion of early encoder features with deeper decoder ones allows UNet models to produce finer-grained dense predictions. While proven in segmentation tasks, the network's benefits are down-weighted for dense regression tasks as these long-range skip connections additionally result in texture transfer artifacts. Specifically for depth estimation, this hurts smoothness and introduces false positive edges which are detrimental to the task due to the depth maps' piece-wise smooth nature. The proposed HybridSkip connections show improved performance in balancing the trade-off between edge preservation, and the minimization of texture transfer artifacts that hurt smoothness. This is achieved by the proper and balanced exchange of information that Hybrid-Skip connections offer between the high and low frequency, encoder and decoder features, respectively.