This hackathon is designed to focus on education, collaboration, and improving ecological data for research and wildlife management. Trail cameras or motion-triggered cameras provide a large amount of invaluable data for these purposes, yet describing each image is time-consuming and requires experts.
Snapshot Wisconsin has collected over one hundred million motion-triggered and site photos from across the state of Wisconsin. While many are classified by species and given population dynamics tags (like age) by volunteers, not all information is converted to a structured format. The goal of this event is to predict some of these population dynamic tags using a labeled dataset supplied by Snapshot Wisconsin, as well as to develop tools that extract countless other tags, features, and descriptors from the images.
Goal Participants must create a useful tool to predict valuable ecological information stored in trail camera images. The leaderboard will be calculated using population dynamics data about deer: * Antlers: Antlered or Antlerless * Age: Adult or Young
For an additional challenge, participants may attempt to extract further information from the photos, such as the time of year or day the photo was taken, whether animals are alert or relaxed, or land cover characteristics. Note that the provided labeled dataset likely does not contain ground truth for these additional predictions; participants may add their own labels, use other datasets, or utilize pretrained models.
Tracks and Awards There is only one Open Track where all submissions follow the same evaluation criteria. The top submission receives the "Top Deer" award and Kudos.
Evaluation Judges will review submissions based on overall usefulness, specifically regarding how easy and effective the tool is for trail camera images. The evaluation rubric (150 points total) considers: * Accuracy: Determined by the F1 score for Antlers and Age predictions. * Additional Predictions: Subjective scoring based on the ecological value, consistency, and accuracy of any extra insights predicted. * Speed: How quickly the tool processes images, scored subjectively based on an average of multiple runs. * Efficiency: A subjective score based primarily on the compute resources required for inference. * Ease of Use: How easy the tool is to set up and use outside of the submission, including maintainability and expandability for research software engineers.