seabirdNET

Repo for using Keras Retinanet for seabird detection in drone imagery

Set up environment

1. Install Docker on Ubuntu (https://docs.docker.com/engine/install/ubuntu/)

2. Install Git

3. Clone this repo

git clone https://github.com/madelinehayes/seabirdNET.git

4. Move into that directory

cd seabirdNET

5. Create a Docker Image and Container based on the Dockerfile in this repo

docker build -t <img_name> Dockerfile
docker run --name <cont_name> -it -p 8888:8888 -p 6006:6006 -v ~/:/host <img_name> 

6. Create a Docker Image and Container for your deep learning environment based on the GPU Dockerfile in this repo

docker build -t <img_name> DockerfileGPU
docker run --name <cont_name> -it -p 8888:8889 -p 6006:6006 -v ~/:/host <img_name>

Your Docker container is now running. Exit that container

exit

To restart your container and attach it to the terminal

docker start <cont_name>
docker attach <cont_name>

Now start jupyter:

jupyter notebook --allow-root --ip 0.0.0.0 /host

7. Now install Keras Retinanet following the instructions here: https://github.com/fizyr/keras-retinanet

• uas_img_handler_FINAL.ipynb is used to split the orthomosaics created from drone imagery into smaller tiles
• The tiles created from uas_img_handler_FINAL.ipynb can be imported into VIA http://www.robots.ox.ac.uk/~vgg/software/via/app/via_image_annotator.html
• via_to_retinanet_FINAL is used to convert the output of VIA into the format Retinanet requires for training data
• albatross_detections_FINAL.ipynb and penguin_detections_FINAL.ipynb trains the model, runs validation and testing, and runs inference on the tiles created from uas_img_handler_FINAL.ipynb
• export_detections_FINAL.ipynb ingests the output from the Retinanet detections and converts into geolocated shapefiles