S010MON/image-locate

Cross View Image Geolocation using Siamese Neural Networks

Image Locate

Cross-view Image Geolocation is the identification of the location of a ground based image by finding a matching aerial photo.

GPU Setup

This project runs within a docker container provided by Tensorflow and requires the use of a GPU. Before running, ensure that you have a GPU compatible with Tensorflow and that it has been set up as per the instructions: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html provided by NVIDIA.

Data Setup

The data used in this study was from the CVUSA dataset from Workman et al. in their paper "Wide-Area Image Geolocalization with Aerial Reference Imagery" (http://arxiv.org/abs/1510.03743). Access to the data can be requested here: https://mvrl.cse.wustl.edu/datasets/cvusa/

To build and run the container, first modify the following line of the docker-compose.yml:

volumes:
  - "./:/tf/notebooks"
  - "/media/your/data/file/:/tf/CVUSA/"    # <== Place your file path to your data here

Your data file should be laid out as follows:

CVUSA/
├── clean_aerial/
│   ├── image_1.jpg
│   ├── image_2.jpg
│   └── image_3.jpg
└── clean_ground/
    ├── image_1.jpg
    ├── image_2.jpg
    └── image_3.jpg

Note that each ground and aerial image have the same name, this allows for the loading of the dataset in proper order. You can include sub-folders at your own risk, however this has caused problems with some of the dataset not loading correctly in the past.

Running the code

To build and run the project, find the docker-compose.yml file and run:

docker-compose up --build

This will build and start a container named image-locate and will provide access to the jupyter server though a link. Training can be run by entering the container and running the trainModel.py file as follows:

docker exec -it image-locate bash

This will enter the container and provide a prompt at the notebooks/ directory.

________                               _______________
___  __/__________________________________  ____/__  /________      __
__  /  _  _ \_  __ \_  ___/  __ \_  ___/_  /_   __  /_  __ \_ | /| / /
_  /   /  __/  / / /(__  )/ /_/ /  /   _  __/   _  / / /_/ /_ |/ |/ /
/_/    \___//_/ /_//____/ \____//_/    /_/      /_/  \____/____/|__/


WARNING: You are running this container as root, which can cause new files in
mounted volumes to be created as the root user on your host machine.

To avoid this, run the container by specifying your user's userid:

$ docker run -u $(id -u):$(id -g) args...

root@1ec13bf1ef31:/tf/notebooks#