Trained a custom YOLOv8 model to identify 9 specific desk objects with 80% training accuracy

I created a custom YOLOv8 object detection model trained on 351 hand-labeled images across 9 desk object classes. The project involved recording 5 training videos from multiple viewpoints (birds-eye, front, horizontal, left, right), extracting frames every 2 seconds, and manually annotating each image. I implemented a complete pipeline including JSON-to-YOLO format conversion, dataset splitting (80% train, 20% validation), and model training for 84 epochs. The final model achieved a fitness score of 0.7944 and successfully detected all target objects in real-world video scenarios after confidence threshold optimization.