ICNet for Real-Time Semantic Segmentation on High-Resolution Images

TL;DR

This paper introduces ICNet, a real-time semantic segmentation network that efficiently processes high-resolution images by using multi-resolution branches and cascade feature fusion, enabling fast inference with good accuracy.

Contribution

The paper presents ICNet, a novel multi-resolution cascade network with feature fusion for real-time semantic segmentation of high-resolution images.

Findings

Real-time inference achieved on a single GPU.

Decent segmentation quality on Cityscapes, CamVid, and COCO-Stuff.

Efficient reduction of computation for pixel-wise labeling.

Abstract

We focus on the challenging task of real-time semantic segmentation in this paper. It finds many practical applications and yet is with fundamental difficulty of reducing a large portion of computation for pixel-wise label inference. We propose an image cascade network (ICNet) that incorporates multi-resolution branches under proper label guidance to address this challenge. We provide in-depth analysis of our framework and introduce the cascade feature fusion unit to quickly achieve high-quality segmentation. Our system yields real-time inference on a single GPU card with decent quality results evaluated on challenging datasets like Cityscapes, CamVid and COCO-Stuff.