Deeply Cascaded U-Net for Multi-Task Image Processing

TL;DR

This paper introduces a novel multi-task neural network architecture based on an extended U-Net with cascading pathways, enabling simultaneous image processing tasks like denoising and segmentation with improved efficiency and performance.

Contribution

The paper presents a deeply cascaded U-Net architecture that integrates multiple image processing tasks into a single model, reducing parameters and enhancing performance over separate or jointly-trained networks.

Findings

Outperforms separate and jointly-trained models in denoising and segmentation

Achieves better results with fewer trainable parameters

Effective in progressive coarse-to-fine semantic segmentation

Abstract

In current practice, many image processing tasks are done sequentially (e.g. denoising, dehazing, followed by semantic segmentation). In this paper, we propose a novel multi-task neural network architecture designed for combining sequential image processing tasks. We extend U-Net by additional decoding pathways for each individual task, and explore deep cascading of outputs and connectivity from one pathway to another. We demonstrate effectiveness of the proposed approach on denoising and semantic segmentation, as well as on progressive coarse-to-fine semantic segmentation, and achieve better performance than multiple individual or jointly-trained networks, with lower number of trainable parameters.