Tiêu đề Soccer Player Re-Identification Assignment ✅

Assignment: Player Re-Identification in Sports Footage Company: Liat.ai Role: AI Intern Submission Format: GitHub Repository or Google Drive Folder Deadline: 1 week Send the submission link (github repo or google drive folder) to [email protected] and [email protected] Overview This assignment assesses your ability to solve real-world computer vision challenges in sports analytics. Your task is to implement a solution for player re-identification — ensuring that the same player retains the same ID, even across different camera feeds or after going out of view. You may choose either of the two tasks described below. All assignment materials are available at the following link: https://drive.google.com/drive/folders/1Nx6H_n0UUI6L-6i8WknXd4Cv2c3VjZTP?usp=s haring Task Options Option 1: Cross-Camera Player Mapping Objective: Given two clips (broadcast.mp4 and tacticam.mp4) of the same gameplay from different camera angles, map the players such that each player retains a consistent ID across both feeds. Instructions: ● Use the provided object detection model to detect players in both videos.
● Match each player from the tacticam video to their corresponding identity in the broadcast video using consistent player_id values. ● You may use any combination of visual, spatial, or temporal features to establish the mapping. Model Download: Object detection model link (for detecting players and the ball): https://drive.google.com/file/d/1-5fOSHOSB9UXyP_enOoZNAMScrePVcMD/view Note: The model is a basic fine-tuned version of Ultralytics YOLOv11, trained specifically on player and ball detection. Option 2: Re-Identification in a Single Feed Objective: Given a 15-second video (15sec_input_720p.mp4), identify each player and ensure that players who go out of frame and reappear are assigned the same identity as before. Instructions: ● Use the provided object detection model to detect players throughout the clip. ● Assign player IDs based on the initial few seconds. ● Maintain the same ID for players when they re-enter the frame later in the video (e.g., near the goal event). ● Your solution should simulate real-time re-identification and player tracking. Model Download: Object detection model link: https://drive.google.com/file/d/1-5fOSHOSB9UXyP_enOoZNAMScrePVcMD/view Note: The model is a basic fine-tuned version of Ultralytics YOLOv11, trained for players and the ball. Submission Requirements Submit your work via a GitHub repository or a Google Drive folder, containing:

You are free to define your own subcomponents and use any publicly available resources. If you have questions, reach out anytime. Good luck, and we’re excited to see what you come up with!