Feature Pyramid Network for Multi-Class Land Segmentation

Selim Seferbekov, Vladimir Iglovikov, Alexander Buslaev, Alexey Shvets; Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2018, pp. 272-275


Semantic segmentation is in-demand in satellite imagery processing. Because of the complex environment, automatic categorization and segmentation of land cover is challenging problem solving which can help to overcome many obstacles in urban planning, environmental engineering or natural landscape monitoring. In this paper we propose an approach for automatic multi-class land segmentation based on fully convolutional neural network of feature pyramid network (FPN) family. This network is consisted of pre-trained on ImageNet Resnet50 encoder and neatly developed decoder. Based on validation results, leader-board score and our own experience this network shows reliable results for the DEEPGLOBE - CVPR 2018 land cover classification sub-challenge. Moreover, this network moderately uses memory that allows to use GTX 1080 or 1080 TI video cards to preform whole training and makes pretty fast predictions.

Related Material

[pdf] [arXiv]
author = {Seferbekov, Selim and Iglovikov, Vladimir and Buslaev, Alexander and Shvets, Alexey},
title = {Feature Pyramid Network for Multi-Class Land Segmentation},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2018}