Syringe exchange programs aim to reduce disease spread among people who use intravenous drugs. The original system relied solely on weight measurements, which could be manipulated. This project develops a computer vision system using DINOv2 embeddings to classify contents and count syringes.
| Task | Model | Performance |
|---|---|---|
| Content classification (3-class) | SVM on DINOv2-base + PCA | 93% accuracy |
| Syringe counting | MLP regression | MAE 1.3 (up to 17 syringes) |
Dataset: 114 annotated images, 70/30 split, augmented via rotation and random cropping.