tasks.csv

task, definition, chain1, chain2, chain3
"Background knowledge text QA", "is a task that asks questions about the basic information of a specific players, referee, team and venues. Ask about those questions could be answer from their WikiPage","Textual Entity Search -> Textual Retrieval Augment -> LLM",,
"Match Situation QA", "is a task that asks questions about the basic information of a specific match, the players and coaches of both teams, and important events of the match (goals, assists, red and yellow cards, etc.) The game range only covers 2014-2024's six european major leagues (Premier, Bundesliga, Serie-a, Ligue-1, Laliga and European Champions League)", "Game Search -> Game Info Retrieval -> Match History Retrieval -> LLM",,
"Match Events and Statistical QA", "is a task to ask questions about the history events of the match. For example, 'How many corners has xxx team get in the first half.'. The game range only covers 2014-2024's six european major leagues (Premier, Bundesliga, Serie-a, Ligue-1, Laliga and European Champions League)", "Game Search -> Game Info Retrieval -> Match History Retrieval -> LLM",,
"Camera Status Classification", "is a task that determines the state of the camera position in the picture at a certain moment in the game.", "Camera Detection -> LLM",,
"Background knowledge Image QA", "is a task that asks questions with one or more images about the basic information of a specific players, referee, team and venues." ,"Entity Recognition -> Textual Entity Search -> Textual Retrieval Augment -> LLM",,
"Jersey Number Recognition", "is a task to identify the jersey numbers of a players in images.", "Number Recognition -> LLM",,
"Score and Time Relevant QA", "is a task asking about questions that starts from scores or gametime, which means you need to recognize the time or score from the given materials of soccer broadcast. Sometimes you need to know game information and sometimes you only need to recognize and then answer the question.", "Score and Time Recognition -> LLM", "Score and Time Recognition -> Game Search ->  Match History Retrieval -> LLM", "Score and Time Recognition -> Game Search -> Game Info Retrieval -> LLM" 
"Game State Relevant QA", "is a task to ask game state questions or tactical questions soccer broadcast (detect people in the image, the position of the players, the formation of the team, etc.)", "Vision Language Model -> LLM",,
"Camera Status Switching", "is a task to judge the state of the camera position switching in the video clip.", "Shot Change -> Camera Detection (twice) -> LLM",,
"Replay Grounding", "is a task to identify which video clip is being replayed from a set of clips, with the first clip serving as the replay.", "Commentary Generation (five times) -> LLM","Commentary Generation (five times) -> Vision Language Model (optional) -> LLM","Replay Grounding -> LLM"
"Action Classification", "is a task to classify the actions of the events on soccer game in the video clip.", "Action Classifier -> LLM",,
"Commentary Generation", "is a task to generate commentary for the events in the video clip.", "Commentary Generation -> LLM",,
"Commentary Relevant QA", "is a task to ask questions about background information of certain player with the question having commentary descriptions.", "Vision Language Model -> LLM","Frame Selection -> Entity Recognition -> Textual Retrieval Augment -> LLM",
"Jersy Color Relevant QA", "is a task to ask questions about soccer stuffs like players, matches. All these questions are with elements of jersey colors.", "Vision Language Model -> LLM","Segment(optional) -> Vision Language Model -> Game Search(optional) -> Textual Retrieval Augment(optional) -> LLM",
"Multi-view Foul Recognition", "is a task to recognize the fouls in the video clip from multiple views.", "Foul Recognition -> LLM",,