oregon critters

A computer vision model for classifying camera trap images from western Oregon -- created by Cara Appel

Please note that this repository is in active development and more documentation will be added. Feel free to contact me to let me know if you are using it.

Intro

This repository contains the trained model weights file oregoncritters_V1.pt, a list of species classes that the model is trained to identify classes.csv, and a folder with several scripts to process images and format predictions for further review.

Requirements

This model uses python scripts and relies on the ultralytics package for deploying the YOLOv8 model. See installation instructions here: https://docs.ultralytics.com/quickstart/#install-ultralytics. Processing can be done on CPU or GPUs.

Workflow overview

1. Rename your images, if necessary. Filenames must be unique within a folder, so rename them if you need to. I like the renaming workflow in the camtrapR package: https://jniedballa.github.io/camtrapR/. Having site and datetime information in the filenames also streamlines analysis.
2. Download or clone this repository to your computer (using the "CODE" button at top right)
3. Create a folder called /data/ within the repository folder and put your images here. Subfolders are okay (e.g., "data/site1", "/data/site2" or even "data/site1/check1")
4. Run the predictions script scripts/1_predict.py
5. Run the formatting script scripts/2_format_predictions.py to generate inputs for various post-processing programs

Usage example

coming soon

Post-processing options

• Njobvu-AI (an open-source browser-based tool): https://github.com/sullichrosu/Njobvu-AI
  • To use Njobvu-AI locally, follow installation instructions from GitHub repo above
  • Then, to create a project in Njobvu-AI with your images, run scripts/3_create_njobvu_project.py
• FiftyOne (a browser-based tool utilizing MongoDB and python): https://docs.voxel51.com/
  • Follow installation instructions from FiftyOne
  • Then, to create a project on FiftyOne with your images, run scripts/4_create_fiftyone_project.py
• Timelapse (a desktop program for review of camera trap images/video): https://saul.cpsc.ucalgary.ca/timelapse/
  • compatibility coming soon