VQA_model/PickScore-main/compute_correlation.py

#!/usr/bin/env python
# -*- coding: utf-8 -*-

"""
计算主观评分和CLIP得分的相关系数（SRCC、KRCC、PLCC）
"""

import os
import sys
import pandas as pd
import numpy as np
from scipy.stats import spearmanr, pearsonr, kendalltau
import argparse
import matplotlib.pyplot as plt

def load_txt_scores(txt_path):
    """从TXT文件加载主观评分"""
    mos_scores = []
    video_paths = []
    
    with open(txt_path, 'r') as f:
        lines = f.readlines()
    
    for line in lines:
        parts = line.strip().split(',')
        if len(parts) >= 2:
            video_path = parts[0]
            try:
                score = float(parts[1])
                video_paths.append(video_path)
                mos_scores.append(score)
            except ValueError:
                print(f"警告: 无法解析分数: {parts[1]}")
    
    return video_paths, mos_scores

def load_csv_scores(csv_path):
    """从CSV文件加载CLIP得分"""
    df = pd.read_csv(csv_path)
    
    # 检查是否包含clip_score列
    if 'clip_score' not in df.columns:
        print(f"错误: CSV文件中没有找到'clip_score'列。可用列: {df.columns.tolist()}")
        sys.exit(1)
    
    video_paths = df['video_path'].tolist()
    clip_scores = df['clip_score'].tolist()
    
    return video_paths, clip_scores

def compute_correlations(mos_scores, clip_scores):
    """计算相关系数"""
    srcc, p_srcc = spearmanr(mos_scores, clip_scores)
    plcc, p_plcc = pearsonr(mos_scores, clip_scores)
    krcc, p_krcc = kendalltau(mos_scores, clip_scores)
    
    return {
        'SRCC': (srcc, p_srcc),
        'PLCC': (plcc, p_plcc),
        'KRCC': (krcc, p_krcc)
    }

def draw_scatter_plot(mos_scores, clip_scores, output_path=None):
    """绘制散点图并计算相关系数"""
    plt.figure(figsize=(10, 6))
    plt.scatter(mos_scores, clip_scores, alpha=0.5)
    
    # 添加线性拟合
    z = np.polyfit(mos_scores, clip_scores, 1)
    p = np.poly1d(z)
    plt.plot(mos_scores, p(mos_scores), "r--", alpha=0.7)
    
    # 计算相关系数
    corrs = compute_correlations(mos_scores, clip_scores)
    
    plt.xlabel('主观评分 (MOS)')
    plt.ylabel('CLIP相似度得分')
    plt.title(f'主观评分与CLIP得分相关性\nSRCC: {corrs["SRCC"][0]:.4f}, PLCC: {corrs["PLCC"][0]:.4f}, KRCC: {corrs["KRCC"][0]:.4f}')
    plt.grid(alpha=0.3)
    
    if output_path:
        plt.savefig(output_path, dpi=300, bbox_inches='tight')
        print(f"散点图已保存至: {output_path}")
    else:
        plt.show()
    
    plt.close()

def main():
    parser = argparse.ArgumentParser(description='计算主观评分和CLIP得分的相关系数')
    parser.add_argument('--txt_path', type=str, required=True, help='TXT文件路径，包含主观评分')
    parser.add_argument('--csv_path', type=str, required=True, help='CSV文件路径，包含CLIP得分')
    parser.add_argument('--output_path', type=str, default=None, help='输出结果的文件路径')
    parser.add_argument('--plot', action='store_true', help='生成散点图')
    args = parser.parse_args()
    
    # 加载数据
    print(f"读取TXT文件: {args.txt_path}")
    txt_videos, mos_scores = load_txt_scores(args.txt_path)
    
    print(f"读取CSV文件: {args.csv_path}")
    csv_videos, clip_scores = load_csv_scores(args.csv_path)
    
    # 检查数据长度
    print(f"TXT文件中的数据数量: {len(txt_videos)}")
    print(f"CSV文件中的数据数量: {len(csv_videos)}")
    
    if len(txt_videos) != len(csv_videos):
        print("警告: TXT和CSV文件中的数据数量不匹配。")
        print("尝试基于视频路径匹配数据...")
        
        # 创建映射表
        matching_data = []
        for i, txt_video in enumerate(txt_videos):
            for j, csv_video in enumerate(csv_videos):
                if txt_video in csv_video or csv_video in txt_video:
                    matching_data.append((mos_scores[i], clip_scores[j]))
                    break
        
        if not matching_data:
            print("错误: 无法匹配TXT和CSV文件中的数据。请确保视频路径兼容。")
            sys.exit(1)
        
        # 提取匹配的分数
        matched_mos, matched_clip = zip(*matching_data)
        mos_scores = matched_mos
        clip_scores = matched_clip
        print(f"找到 {len(matching_data)} 对匹配的数据。")
    
    # 计算相关系数
    corrs = compute_correlations(mos_scores, clip_scores)
    
    # 打印结果
    print("\n相关系数分析结果:")
    print(f"SRCC: {corrs['SRCC'][0]:.4f} (p-value: {corrs['SRCC'][1]:.4f})")
    print(f"PLCC: {corrs['PLCC'][0]:.4f} (p-value: {corrs['PLCC'][1]:.4f})")
    print(f"KRCC: {corrs['KRCC'][0]:.4f} (p-value: {corrs['KRCC'][1]:.4f})")
    
    # 保存结果
    if args.output_path:
        with open(args.output_path, 'w') as f:
            f.write(f"总数据点: {len(mos_scores)}\n")
            f.write(f"SRCC: {corrs['SRCC'][0]:.4f} (p-value: {corrs['SRCC'][1]:.4f})\n")
            f.write(f"PLCC: {corrs['PLCC'][0]:.4f} (p-value: {corrs['PLCC'][1]:.4f})\n")
            f.write(f"KRCC: {corrs['KRCC'][0]:.4f} (p-value: {corrs['KRCC'][1]:.4f})\n")
        print(f"结果已保存至: {args.output_path}")
    
    # 生成散点图
    if args.plot:
        plot_path = args.output_path.replace('.txt', '.png') if args.output_path else 'correlation_plot.png'
        draw_scatter_plot(mos_scores, clip_scores, plot_path)

if __name__ == "__main__":
    main()