I need an end-to-end computer-vision pipeline that can take raw tennis match footage and automatically 1) detect every player, ball, and racket on screen, and 2) classify what is happening at each moment. The pipeline must recognise player actions, tell different shot types apart, and keep consistent track of each player’s movement throughout the rally. Also with that information it must provide the strengths and tips to improve the movement and score from 0 to 100 depending of the quality of the shot. You are free to choose the stack you are most productive with—TensorFlow, PyTorch, YOLOv5/8, Detectron2, OpenCV, Deep SORT or ByteTrack are all perfectly acceptable as long as the end result is accurate and reproducible. I will supply a representative sample of matches for training and evaluation, and can label additional clips if the model needs more data. The system should ingest standard MP4 files, and produce: Build a detection and classification pipeline using: • Roboflow + YOLO, or • Ultralytics YOLOv8/YOLO11 + MediaPipe, or • MoveNet/SensiAI + classifier • Detect: player, racket, ball, pose, shot type. • Compute timing and technical metrics. • Generate structured JSON: "type_of_shot": "bandeja", "strengths": [], "improvements": [], "score": 82, "overlay_url": "" • Generate human-like feedback using GPT-4o or simirlar. • Benchmark latency + cost per video. • Deliver API or script ready for integration. REQUIRED SKILLS • Computer vision (YOLO, pose estimation, object tracking) • Python, OpenCV If you have previous sports-analytics or pose-estimation experience, please mention it—otherwise, strong object-detection and sequence-classification results will speak for themselves. I am ready to start as soon as we align on milestones and dataset access.