The core detection model is based on a custom-trained YOLO architecture capable of real-time inference and annotation. app/ ├── static/ │ ├── uploads/ │ │ ├── videos/ # Uploaded raw videos │ │ └── ...