/Multi-connection-attention-networks

Multi-connection attention networks

Primary LanguagePython

Multi-connection attention network

image label

The pretrained model of Potsdam Dataset will be available in a month.

by Jicheng WANG, Li SHEN, Wenfan QIAO, Yanshuai DAI, Zhilin LI, details (paper).

Introduction

This code is the implementation of SWJ_2 in ISPRS Potsdam labeling challenge 2D. This network mainly consist of two module, i.e., multi-connection resnet and class-specific attention model.

Installation

For installation, please follow the instructions of tensorflow.Both GPU and CPU are compatible. Noted that the cuDNN is needed for GPU version. The version of tensorflow tested is 1.10.0.

Usage

  1. Clone this repository
git clone https://github.com/WindWang2/Multi-connection-attention-networks.git
  1. Download the pretrained model and prepare the images
  2. Change the code of test.py (path of pretrained model and directory of test images)
  3. run the code
python3 test.py

train.py can be modified to train the model.

Update

...