First-Impression-llm-version/example_video_native.py

"""
示例：使用Qwen2.5-VL原生视频输入进行人格分析
"""
import os
from model_llm import PersonalityAnalyzer
from config import Config

def main():
    # 初始化分析器
    analyzer = PersonalityAnalyzer()
    
    # 示例视频路径（请替换为实际路径）
    video_path = "/path/to/your/video.mp4"
    
    # 检查视频文件是否存在
    if not os.path.exists(video_path):
        print(f"错误: 视频文件不存在: {video_path}")
        print("请替换为实际的视频文件路径")
        return
    
    print("=" * 60)
    print("使用Qwen原生视频输入进行人格分析")
    print("=" * 60)
    
    # 方法1: 使用原生视频输入（推荐）
    print("\n【方法1】使用原生视频输入（use_native_video=True）")
    result1 = analyzer.predict(
        video_path=video_path,
        use_native_video=True,  # 使用原生视频输入
        max_frames=20
    )
    
    print("\n预测结果:")
    print(f"外向性 (Extraversion): {result1['scores']['Extraversion']:.3f}")
    print(f"神经质 (Neuroticism): {result1['scores']['Neuroticism']:.3f}")
    print(f"宜人性 (Agreeableness): {result1['scores']['Agreeableness']:.3f}")
    print(f"尽责性 (Conscientiousness): {result1['scores']['Conscientiousness']:.3f}")
    print(f"开放性 (Openness): {result1['scores']['Openness']:.3f}")
    print(f"\n推理过程: {result1.get('reasoning', 'N/A')}")
    
    # 方法2: 使用帧提取模式（如果需要更多控制）
    print("\n" + "=" * 60)
    print("【方法2】使用帧提取模式（use_native_video=False）")
    result2 = analyzer.predict(
        video_path=video_path,
        use_native_video=False,  # 提取帧
        max_frames=20,
        frame_selection="smart"  # 智能帧选择
    )
    
    print("\n预测结果:")
    print(f"外向性 (Extraversion): {result2['scores']['Extraversion']:.3f}")
    print(f"神经质 (Neuroticism): {result2['scores']['Neuroticism']:.3f}")
    print(f"宜人性 (Agreeableness): {result2['scores']['Agreeableness']:.3f}")
    print(f"尽责性 (Conscientiousness): {result2['scores']['Conscientiousness']:.3f}")
    print(f"开放性 (Openness): {result2['scores']['Openness']:.3f}")
    print(f"\n推理过程: {result2.get('reasoning', 'N/A')}")
    
    # 方法3: 多模态输入（视频 + 文本 + 音频）
    print("\n" + "=" * 60)
    print("【方法3】多模态输入（视频 + 文本 + 音频）")
    
    # 假设有文本转录和音频文件
    text_path = video_path.replace(".mp4", "_transcription.txt")
    audio_path = video_path.replace(".mp4", ".wav")
    
    result3 = analyzer.predict(
        text=text_path if os.path.exists(text_path) else None,
        video_path=video_path,
        audio_path=audio_path if os.path.exists(audio_path) else None,
        use_native_video=True
    )
    
    print("\n预测结果:")
    print(f"外向性 (Extraversion): {result3['scores']['Extraversion']:.3f}")
    print(f"神经质 (Neuroticism): {result3['scores']['Neuroticism']:.3f}")
    print(f"宜人性 (Agreeableness): {result3['scores']['Agreeableness']:.3f}")
    print(f"尽责性 (Conscientiousness): {result3['scores']['Conscientiousness']:.3f}")
    print(f"开放性 (Openness): {result3['scores']['Openness']:.3f}")
    print(f"\n推理过程: {result3.get('reasoning', 'N/A')}")

if __name__ == "__main__":
    main()