I need a YOLO-based solution that can reliably pick out humans, vehicles, and animals in real-time video streams or still images. The exact deployment setting may vary—sometimes indoors, sometimes outdoors—so the model should generalise well to differing light conditions, backgrounds, and camera angles. What I expect from you: • A trained YOLO model (v5, v7, v8, or the version you feel offers the best accuracy–speed balance) fine-tuned on datasets rich in the three object classes above. • An easy-to-run inference script written in Python that accepts an RTSP feed, webcam, or folder of images and returns bounding boxes, class labels, and confidence scores. • Clear instructions for reproducing the training and for exporting the weights so I can retrain or extend the model later if needed. • A short performance report: mAP on a held-out validation set, FPS on 1080p video, and recommended hardware specs. OpenCV, PyTorch, and any other libraries you consider essential are fine—just document the environment so setup is painless. Let me know in your bid which YOLO flavour you prefer and why.