Peregrine

Detect and classify bird species in realtime on a video feed

Building

HALO

You'll need the Tensyr HALO SDK. Make sure it's installed in /opt/tensyr (if you have a tarball release, just untar it there)

We use CMake for our build. To build Peregrine, first create and cd to a build directory and then run:

cmake path/to/sources
make
make install

This will create a new install directory inside of your build directory with all of Peregrine's runnable artificats.

Depedencies

Peregrine depends on opencv and ffmpeg. You can either install these or use the docker image (runtime.df). If using the docker image, you can run enter-container.sh to run make inside the image

Running

Once you have an install directory, just run the run script inside of the install directory (make sure to cd to the install directory first). This will take in a video (with the -f flag) and output to a file called video_out.mp4

Docker

HALO only supports Ubuntu 16.04, but you can use the provided dockerfile to run Peregrine on any linux system. Build the dockerfile (runtime.df) and point the env to a folder with the HALO debian. I.e.:

  docker build -f runtime.df /path/to/halo/deb

The run script supports running Peregrine inside a docker container. You'll need to specify an image via the -i flag.

Dataset

We use the NABirds V1 dataset (DOI: 10.1109/CVPR.2015.7298658) for species classification training, generously available through the Cornell Lab of O.