import pandas as pd
import gradio as gr
import csv
import json
import os
import shutil
from huggingface_hub import Repository

HF_TOKEN = os.environ.get("HUGGINGFACE_TOKEN")

MODEL_INFO = [
    "Model",
    "Avg",
    "Visual Quality",
    "Temporal Consistency",
    "Dynamic Degree",
    "Text-to-Video Alignment",
    "Factual Consistency"
    ]

DATA_TITILE_TYPE = ['markdown', 'number', 'number', 'number', 'number', 'number',]

SUBMISSION_NAME = "VideoScore-Leaderboard"
SUBMISSION_URL = os.path.join("https://huggingface.co/datasets/hexuan21/", SUBMISSION_NAME)
CSV_DIR = "./VideoScore-Leaderboard/leaderboard_res.csv"

COLUMN_NAMES = MODEL_INFO

LEADERBORAD_INTRODUCTION = """# VideoScore Leaderboard

    🏆 Welcome to the **VideoScore Leaderboard**! The leaderboard covers many popular text-to-video generative models and evaluates them on 5 dimensions: <br> 
    
    "Visual Quality", "Temporal Consistency", "Dynamic Degree", "Text-to-Video Alignment", "Factual Consistency"
    
    We sample 200 prompts from <a href="https://arxiv.org/abs/2403.06098">VidProM</a> to generate 200 videos using various T2V models (for those closed-source model, we generate 100).    

    <a href='https://hits.seeyoufarm.com'><img src='https://hits.seeyoufarm.com/api/count/incr/badge.svg?url=https%3A%2F%2Fhuggingface.co%2Fspaces%2FTIGER-Lab%2FVideoScore-Leaderboard&count_bg=%23C7C83D&title_bg=%23555555&icon=&icon_color=%23E7E7E7&title=hits&edge_flat=false'></a>
    """

TABLE_INTRODUCTION = """
    """

LEADERBORAD_INFO = """
"""

CITATION_BUTTON_LABEL = "Copy the following snippet to cite the t2v models and the used metrics"
CITATION_BUTTON_TEXT = r"""


"""

def get_df():
    repo = Repository(local_dir=SUBMISSION_NAME, clone_from=SUBMISSION_URL, use_auth_token=HF_TOKEN)
    repo.git_pull()
    df = pd.read_csv(CSV_DIR)
    df['Model'] = df['Model'].apply(lambda x: f"[{x.split(']')[0][1:]}]({x.split('(')[1][:-1]})")
    df['Avg'] = df[["Visual Quality",
    "Temporal Consistency",
    "Dynamic Degree",
    "Text-to-Video Alignment",
    "Factual Consistency"]].mean(axis=1).round(2)
    df = df.sort_values(by=['Avg'], ascending=False)
    return df[COLUMN_NAMES]


def refresh_data():
    return get_df()