Cowspiracy

Overview

Cowspiracy uses computer vision to automatically detect and count animals (cattle, goats, camels) in large collections of field images. The goal is to quantify grazing pressure as a driver of soil erosion and land degradation in two regions: the Sahel (Senegal) and the Gobi Desert.

Partners

My Role

Lead the data science component, working directly with ecologists at iEES Paris and IRD. Responsible for detection pipeline design, modelling strategy and evaluation. Implement detection YOLO-based models.

Technical Approach

• Detection model: fine-tuned YOLO-based object detection on annotated field images
• Training data: manually annotated images from field campaigns in Senegal
• Scalability: batch inference over thousands of images per campaign
• Outputs: animal counts, temporal trends

Stack: Python · PyTorch · Ultralytics YOLO

Scientific Context

Automated detection tools allow researchers to move from labour-intensive manual counting to systematic, reproducible landscape-scale monitoring — enabling long-term ecological studies on desertification that were previously infeasible.

Status

🟡 In progress