Convolutional Patch Networks with Spatial Prior for Road Detection and Urban Scene Understanding

TL;DR

This paper introduces convolutional patch networks with spatial priors for pixel-wise classification, achieving state-of-the-art results in road detection and urban scene understanding by combining appearance and spatial information.

Contribution

It presents a novel convolutional patch network architecture that incorporates spatial priors for improved pixel-wise labeling in scene understanding tasks.

Findings

Achieved state-of-the-art results on KITTI and LabelMeFacade datasets.

Demonstrated the effectiveness of spatial priors in convolutional networks.

Provided guidelines for training convolutional networks on image patches.

Abstract

Classifying single image patches is important in many different applications, such as road detection or scene understanding. In this paper, we present convolutional patch networks, which are convolutional networks learned to distinguish different image patches and which can be used for pixel-wise labeling. We also show how to incorporate spatial information of the patch as an input to the network, which allows for learning spatial priors for certain categories jointly with an appearance model. In particular, we focus on road detection and urban scene understanding, two application areas where we are able to achieve state-of-the-art results on the KITTI as well as on the LabelMeFacade dataset. Furthermore, our paper offers a guideline for people working in the area and desperately wandering through all the painstaking details that render training CNs on image patches extremely difficult.