Skip to content
master
Switch branches/tags
Code

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
Oct 11, 2018
May 26, 2021

Overview

This repo contains the tools for training, running, and evaluating detectors and classifiers for images collected from motion-triggered camera traps. The core functionality provided is:

  • Data parsing from frequently-used camera trap metadata formats into a common format
  • Training and evaluation of detectors, particularly our "MegaDetector", which does a pretty good job finding terrestrial animals in a variety of ecosystems
  • Training and evaluation of species-level classifiers for specific data sets
  • A Web-based demo that runs our models via a REST API that hosts them on a Web endpoint
  • Miscellaneous useful tools for manipulating camera trap data
  • Research experiments we're doing around camera trap data (i.e., some directories are highly experimental and you should take them with a grain of salt)

Classifiers and detectors are trained using TensorFlow.

This repo is maintained by folks in the Microsoft AI for Earth program who like looking at pictures of animals. I mean, we want to use machine learning to support conservation too, but we also really like looking at pictures of animals.

Who is using the AI for Earth camera trap tools?

We work with ecologists all over the world to help them spend less time annotating images and more time thinking about conservation. You can read a little more about how this works on our AI for Earth camera trap collaborations page.

You can also read about what we do to support camera trap researchers in our recent blog post.

Here are a few of the organizations that have used AI for Earth camera trap tools... we're only listing organizations who (a) we know about and (b) have generously allowed us to refer to them here, so if you're using MegaDetector or other tools from this repo and would like to be added to this list, email us!

  • Idaho Department of Fish and Game
  • San Diego Zoo Global
  • University of Washington Quantitative Ecology Lab
  • University of Idaho
  • Borderlands Research Institute at Sul Ross State University
  • Borneo Nature Foundation
  • Parks Canada
  • Australian Wildlife Conservancy
  • Lab of Dr. Bilal Habib at the Wildlife Institute of India
  • Royal Society for the Protection of Birds (RSPB)
  • Wildlife Protection Solutions
  • Island Conservation
  • Synthetaic
  • School of Natural Sciences, University of Tasmania
  • Arizona Department of Environmental Quality
  • Wildlife Research, Oregon Department of Fish and Wildlife
  • National Wildlife Refuge System, Southwest Region, US Fish and Wildlife
  • Mammal Spatial Ecology and Conservation Lab at Washington State University
  • Point No Point Treaty Council
  • SPEA (Portuguese Society for the Study of Birds)
  • Ghost Cat Analytics
  • EcoLogic Consultants Ltd.
  • Smithsonian Northern Great Plains Program
  • Federal University of Amapá, Ecology and Conservation of Amazonian Vertebrates Research Group
  • Hamaarag, The Steinhardt Museum of Natural History, Tel Aviv University
  • Czech University of Life Sciences Prague
  • Ramat Hanadiv Nature Park, Israel
  • TU Berlin, Department of Ecology
  • DC Cat Count, led by the Humane Rescue Alliance
  • Center for Biodiversity and Conservation at the American Museum of Natural History
  • Camelot
  • Graeme Shannon's Research Group at Bangor University
  • Snapshot USA
  • University of British Columbia Wildlife Coexistence Lab
  • Michigan Department of Natural Resources, Wildlife Division

Data

This repo does not directly host camera trap data, but we work with our collaborators to make data and annotations available whenever possible on lila.science.

Models

This repo does not extensively host species classification models, though we will release models when they are at a level of generality that they might be useful to other people. But...

MegaDetector

Speaking of models that might be useful to other people, we have trained a one-class animal detector trained on several hundred thousand bounding boxes from a variety of ecosystems. Lots more information – including download links – on the MegaDetector page.

Here's a "teaser" image of what detector output looks like:

Red bounding box on fox

Image credit University of Washington.

Contact

For questions about this repo, contact cameratraps@microsoft.com.

Contents

This repo is organized into the following folders...

api

Code for hosting our models as an API, either for synchronous operation (e.g. for real-time inference or for our Web-based demo) or as a batch process (for large biodiversity surveys).

classification

Code for training species classifiers on new data sets, generally trained on crops generated via an existing detector. We'll release some classifiers soon, but more importantly, here's a tutorial on training your own classifier using our detector and our training pipeline.

Oh, and here's another "teaser image" of what you get at the end of training a classifier:

data_management

Code for:

  • Converting frequently-used metadata formats to COCO Camera Traps format
  • Creating, visualizing, and editing COCO Camera Traps .json databases
  • Generating tfrecords

demo

Source for the Web-based demo of our MegaDetector model (we'll release the demo soon!).

detection

Code for training and evaluating detectors.

research

Ongoing research projects that use this repository in one way or another; as of the time I'm editing this README, there are projects in this folder around active learning and the use of simulated environments for training data augmentation.

sandbox

Random things that don't fit in any other directory. Currently contains a single file, a not-super-useful but super-duper-satisfying and mostly-successful attempt to use OCR to pull metadata out of image pixels in a fairly generic way, to handle those pesky cases when image metadata is lost.

Installation

We use conda to manage our Python package dependencies. Conda is a package and environment management system. You can install a lightweight distribution of conda (Miniconda) for your OS via installers at https://docs.conda.io/en/latest/miniconda.html.

Initial setup

Utility and visualization scripts

The required Python packages for running utility and visualization scripts in this repo are listed in environment.yml. To set up your environment for these scripts, in your shell, navigate to the root directory of this repo and issue the following command to create a virtual environment via conda called cameratraps (specified in the environment file) and install the required packages:

conda env create --file environment.yml

For unix users, you need to have gcc installed in order to compile the pip packages. If you do not already have gcc installed, run the following command before creating the conda environment:

sudo apt update
sudo apt install build-essential

Machine learning scripts

Scripts that execute machine learning code – specifically, scripts in the folders api, detection, and classification – require additional depdendencies. In particular, the detection/run_tf_detector*.py scripts should use environment-detector.yml to set up the environment, as follows:

conda env create --file environment-detector.yml

This environment file allows any TensorFlow version from 1.9 to 1.15 to be installed, but you may need to adjust that version for your environment. Specifically, if you are running on an Azure Data Science Virtual Machine (which has CUDA 10.1 as of the time I'm writing this), you may receive a CUDA error, in which case you should change the line:

- tensorflow-gpu>=1.9.0, <1.15.0

...to:

- tensorflow-gpu=1.13.1

...before creating your environment.

Troubleshooting

If you run into an error while creating either of the above environments, try updating conda to version 4.5.11 or above. Check the version of conda using conda --version.

Usage

To enter the conda virtual environment at your current shell, run:

conda activate cameratraps

...or, if you used the environment-detector.yml file above:

conda activate cameratraps-detector

You should see (cameratraps) prepended to the command line prompt. Invoking python or jupyter notebook will now be using the interpreter and packages available in this virtual env.

To exit the virtual env, issue conda deactivate.

Add additional packages

If you need to use additional packages, add them to the environment file and run

conda env update --name cameratraps --file environment.yml --prune

or

conda env update --name cameratraps-detector --file environment-detector.yml --prune

Other notes

In some scripts, we also assume that you have the AI for Earth utilities repo (ai4eutils) cloned and its path appended to PYTHONPATH. You can append a path to PYTHONPATH for the current shell session by executing the following on Windows:

set PYTHONPATH="%PYTHONPATH%;c:\wherever_you_put_the_ai4eutils_repo"

You can do this with the following on Linux:

export PYTHONPATH="$PYTHONPATH:/absolute/path/to/repo/ai4eutils"

Adding this line to your ~/.bashrc (on Linux) modifies PYTHONPATH permanently.

We also do our best to follow Google's Python Style Guide, and we have adopted their pylintrc file, with the following differences:

  • indent code blocks with 4 spaces (instead of 2)

To lint a file, run pylint with the CameraTraps repo folder as the current working directory. This allows pylint to recognize the pylintrc file. For example,

pylint classification/train_classifier.py

Gratuitous pretty camera trap picture

Bird flying above water

Image credit USDA, from the NACTI data set.

Contributing

This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit cla.microsoft.com.

When you submit a pull request, a CLA-bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., label, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.

This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.

License

This repository is licensed with the MIT license.