Iterative Deep Learning for Road Topology Extraction

Fork me on GitHub

Universitat Oberta de Catalunya

ETH Zürich

introduction

Paper

This paper tackles the task of estimating the topology of road networks from aerial images. Building on top of a global model that performs a dense semantical classification of the pixels of the image, we design a Convolutional Neural Network (CNN) that predicts the local connectivity among the central pixel of an input patch and its border points. By iterating this local connectivity we sweep the whole image and infer the global topology of the road network, inspired by a human delineating a complex network with the tip of their finger. We perform an extensive and comprehensive qualitative and quantitative evaluation on the road network estimation task, and show that our method also generalizes well when moving to networks of retinal vessels.

If you find this work useful, please consider citing:

Carles Ventura, Jordi Pont-Tuset, Sergi Caelles, Kevis-Kokitsi Maninis, Luc Van Gool, "Iterative Deep Learning for Road Topology Extraction" in British Machine Vision Conference (BMVC) 2018.

Download our paper on arXiv.

Results

We provide quantitative results on two datasets (road and retinal images).

Massachusetts Roads Dataset (road images)

MRD results

DRIVE dataset (retinal images)

DRIVE results

Examples

We provide examples of iterative predicted network topology for two different datasets (road and retinal images).

Massachusetts Roads Dataset (road images)

MRD results

DRIVE dataset (retinal images)

DRIVE results

code

pytorch

We implement our models using Pytorch.

Find source code on github.