DDCM Network for Semantic Mapping of Remote Sensing Images

Qinghui Liu, Michael Kampffmeyer,

Robert Jenssen, and Arnt-Børre Salberg

May. 9, 2019

Introduction

Semantic Mapping

ISPRS Vaihingen IR-RG Image :

http://www2.isprs.org/commissions/comm3/wg4/2d-sem-label-vaihingen.html

semantic segmentation / scene parsing / pixel-wise classification

Methods

The DDCM architecture composed of N DC modules with various dilation rates { 1, 2, 3, ... , N }.

End-to.end pipeline of DDCM-Net for semantic mapping of remote sensing images.

DDCM-ResNet50 Details with 3 x 256 x 256 Input Size

Results

Public Remote Sensing Datasets 

ISPRS Vaihingen: http://www2.isprs.org/commissions/comm3/wg4/2d-sem-label-vaihingen.html

ISPRS Potsdam: http://www2.isprs.org/commissions/comm3/wg4/2d-sem-label-potsdam.html

Comparison: Parameters Size, GFLOPs and mIoU

(-1.5%)

(+6x)

(+3x)

(GFLOPs measured on input size 3 x 256 x256),

Potsdam hold-out (14 unseen images) test results.

Vaihingen hold-out (17 unseen images) test results.

Test Image

Predition

GT

Potsdam RGB, 5cm

Test Image

Predition

GT

Vaihingen IR-RG

Thanks @ Merci...

Conclusions

  • Fast and Accurate
  • Simple and Robust
  • Flexible and Efficent

Thank You

Questions?

nldl2019-ddcm

By Brian Liu

nldl2019-ddcm

NLDL2019 talk

  • 1,468