This repository provides the link to our introduced large-scale condition changing datasets, as well as scripts to extract condition-robust features using models from the paper "Deep learning features at scale for visual place recognition" published by Zetao Chen, et al. on ICRA 2017.
- Ubuntu
- Follow this guide to install Caffe library on Ubuntu
-
The currently released dataset version can be downloaded here
-
Download the "HybridNet" model from https://goo.gl/kF6nQh and copy them to the folder "HybridNet" in this repository, or download the "AmosNet" model from https://goo.gl/6xtjwD and copy them to the folder "AmosNet";
-
You will need to update the file 'nordland.txt' using the file pathes of your images. Each line in this file specifies one image whose feature is to be extracted.
-
The file 'extract_feat_usingAMOS.py' extract features from the fc7 layer. Update it if you need to extract features from other layers.
-
Run the 'extract_feat.sh'